Giulia Manca di
a,
⇑
, Alicja Mikolajczyk
b,m,
⇑
, Vigneshwa i K. Annapoo ani
c
,
Aileen Bahl
d
, Kos as Blekos
e
, Jaanus Bu k
, Ya kın A. Çe in
g
,
Kons an inos Chai e akis
e
, Su apa Du a
c
, Lau a Esco ihuela
g
, Ka olina Jagiello
b,m
,
Ankush Singhal
h
, Rianne an de Pol
h
, Miguel A. Baña es
i
, Nicolae-Vio el Buche e
c
,
Monica Cala ayud
j
, Ve ónica I. Dumi
d
, Da ide Ga dini
k
, Nina Jeliazko a
l
,
And ea Haase
d
,E fie Ma coulaki
e
, Benjamí Ma o ell
g
, Tomasz Puzyn
b,m
,
G.J. Agu Se ink
h
, Felice C. Simeone
k
, Kaido Tämm
, Eliodo o Chia azzo
a,
⇑
a
Poli ecnico di To ino, Co so Duca degli Ab uzzi, 24, To ino 10129, I aly
b
Labo a o y o En i onmen al Chemome ics, Ins i u e o En i onmen al and Human Heal h P o ec ion, Facul y o Chemis y, Uni e si y o Gdansk, 80-308
Gdansk, Poland
c
School o Physics & Ins i u e o Disco e y, Uni e si y College Dublin, Belfield, Dublin 4, I eland
d
Ge man Fede al Ins i u e o Risk Assessmen (B R), Depa men o Chemical and P oduc Sa e y, Be lin, Ge many
e
Ins i u e o Nuclea & Radiological Sciences & Technology, Ene gy & Sa e y, Na ional Cen e o Scien ific Resea ch 
nDemok i os_
z, A hens, G eece
Ins i u e o Chemis y, Uni e si y o Ta u, Ra ila 14a, 50411 Ta u, Es onia
g
Depa amen d’Enginye ia Química, Uni e si a Ro i a i Vi gili, A inguda dels Països Ca alans, 26, 43007 Ta agona, Spain
h
Leiden Ins i u e o Chemis y, Leiden Uni e si y, P.O. Box 9502, 2300 RA Leiden, The Ne he lands
i
Ins i u e o Ca alysis, ICP-CSIC, Ma ie Cu ie 2, E-28049 Mad id, Spain
j
Labo a oi e de Chimie Theo ique, CNRS, So bonne Uni e si é, 4 Place Jussieu, 75005 Pa is, F ance
k
ISTEC-CNR, Ins i u e o Science and Technology o Ce amics - Na ional Resea ch Council o I aly, Faenza, I aly
l
Ideaconsul L d, Sofia, Bulga ia
m
QSAR Lab L d., T zy Lipy 3, 80-172 Gdansk, Poland
In ecen yea s, an inc easing numbe o di e se Enginee ed Nano-Ma e ials (ENMs), such as
nanopa icles and nano ubes, ha e been included in many echnological applica ions and consume
p oduc s. The desi able and unique p ope ies o ENMs a e accompanied by po en ial haza ds whose
impac s a e di ficul o p edic ei he quali a i ely o in a quan i a i e and p edic i e manne .
Alongside es ablished me hods o expe imen al and compu a ional cha ac e isa ion, physics-based
modelling ools like molecula dynamics a e inc easingly conside ed in Sa e and Sus ainabili y-by-
design (SSbD) s a egies ha pu use heal h and en i onmen al impac a he cen e o he design and
de elopmen o new p oduc s. Hence, he u he de elopmen o such ools can suppo sa e and
sus ainable inno a ion and i s egula ion.
This pape s ems om a communi y e o and p esen s he ou come o a ou -yea -long discussion
on he benefi s, capabili ies and limi a ions o adop ing physics-based modelling o compu ing
sui able ea u es o nanoma e ials ha can be used o oxici y assessmen o nanoma e ials in
combina ion wi h da a-based models and expe imen al assessmen o oxici y endpoin s. We e iew
A compu a ional iew on nanoma e ial
in insic and ex insic ea u es o
nanosa e y and sus ainabili y
⇑
Co esponding au ho s.
E-mail add esses: Manca di, G. ([email p o ec ed]), Mikolajczyk, A. ([email p o ec ed]), Chia azzo, E. ([email p o ec ed]).
344
1369-7021/Ó2023 Published by Else ie L d. This is an open access a icle unde he CC BY-NC-ND license (h p://c ea i ecommons.o g/licenses/by-nc-nd/4.0/).
Ma e ials Today dVolume 67 dJuly/Augus 2023 RESEARCH
RESEARCH: Re iew
mode n mul iscale physics-based models ha gene a e ad anced sys em-dependen (in insic) o ime-
and en i onmen -dependen (ex insic) desc ip o s/ ea u es o ENMs (p ima ily, bu no limi ed o
nanopa icles, NPs), wi h he o me being ela ed o he ba e NPs and he la e o hei dynamic
finge p in ing upon en e ing biological media. The ocus is on (i) e ec i ely ep esen ing all
nanopa icle a ibu es o mul icomponen nanoma e ials, (ii) gene a ion and inclusion o in insic
nano o m p ope ies, (iii) inclusion o selec ed ex insic p ope ies, (i ) he necessi y o conside ing
dis ibu ions o s uc u al ad anced ea u es a he han only a e ages. This e iew enables us o
iden i y and highligh a numbe o key challenges associa ed wi h ENMs’da a gene a ion, cu a ion,
ep esen a ion and use wi hin machine lea ning o o he ad anced da a-d i en models o ul ima ely
enhance oxici y assessmen . Finally, he se up o dedica ed da abases as well as he de elopmen o
g ouping and ead-ac oss s a egies based on he mode o ac ion o ENMs using omics me hods a e
iden ified as eme ging me hodologies o sa e y assessmen and educ ion o animal es ing.
Keywo ds: Nanoin o ma ics; Nanosa e y; Enginee ed nanoma e ials; Physicochemical desc ip o s; Ma e ials modeling;
Machine lea ning; G ouping app oaches; Mul iscale modeling; Sa e and Sus ainabili y-by-design(SSbD)
In oduc ion
ENMs pe o m key dedica ed asks in ca alysis [1], medicine [2,3],
ag icul u e [4], ood [5] and ene gy [6,7] among many o he s,
because o hei excep ional and o en unique cha ac e is ics.
ENMs unde pin new p oduc s and de ices equi ing con ol o
ma e a he nanome e scale, and he possibili y o fine- une
hei syn hesis p o ocols esul s in coun less a ian s wi h di e -
en physicochemical p ope ies. The excep ional p ope ies o
ENMs - s emming om he manipula ion o ma e a he a omic
scale - can also come wi h ye li le-known isks o human heal h
and he en i onmen . In ac , he biological ac i i y o ENMs is
hough o be closely ela ed o hei physicochemical cha ac e -
is ics, which may be al e ed by he biological medium i sel du -
ing he li e ime o ENMs (e.g., p o ein co ona o ma ion and
s uc u al modifica ions). Bo h he in insic and ex insic physic-
ochemical ea u es o ENMs a e key o ecognizing and p edic -
ing haza ds as well as assessing sa e y cha ac e is ics. The o me
a e ela ed o chemical composi ion, c ys al s uc u e, size, shape
and su ace s uc u e, whe eas he la e pe ain o he non- i ial
in e ac ion wi h he en i onmen . As such, a comple e se o
hose ea u es (he e e e ed o as ad anced desc ip o s) a e c i ical
in nanoin o ma ics o da a-based model de elopmen , such as
he popula quan i a i e nanos uc u e–ac i i y/p ope y models
(QSAR/QSPR). In his espec , p edic i e models o e unp ece-
den ed oppo uni ies o knowledge-based op imiza ion and
de elopmen o new ENMs imp o ing hei unc ionali y and -
a he same ime - minimizing unexpec ed heal h and/o en i-
onmen al isks. P elimina y in silico sc eening o possible e -
sions o new ENMs can hus lead o op imal nanos uc u es
wi h educed haza dous cha ac e is ics e en be o e he p oduc-
ion s age. Howe e , he pace o u he p og ess in nano echnol-
ogy will c i ically depend on a syne gis ic knowledge in eg a ion
o expe imen al e idence wi h da a om eliable heo e ical and
compu a ional models.
The combined s udy o ma e ials modeling (nanos uc u e
cha ac e iza ion) and p edic i e models o he design o sa e
Nomencla u e
AIMD ab ini ioMolecula Dynamics
AO(P) Ad e se Ou come (Pa hway)
API Applica ion P og amming In e ace
BD B ownian Dynamics
CG(MD) Coa se-G ained (Molecula Dynamics)
CSS Chemicals S a egy o Sus ainabili y
DFT Densi y Func ional Theo y
DFTB Densi y Func ional Tigh -Binding
DLVO De jaguin-Landau-Ve wey-O e beek
ENM Enginee ed NanoMa e ials
FAIR Findable, Accessible, In e ope able and Reusable
GA Gene ic Algo i hm
GAN Gene a i e Ad e sa ial Ne wo k
HOMO Highes Occupied Molecula O bi al
hPF hyb id Pa icle-Field
KE Key E en s
LCA Li e Cycle Assessmen
LDM Liquid D op Model
LUMO Lowes Unoccupied Molecula O bi al
MD Molecula Dynamics
MDReaxFF Reac i e Fo ce Field Molecula Dynamics
MIE Molecula Ini ia ing E en s
ML Machine Lea ning
MoA Mode o Ac ion
MODA MOdelling DA a fiche
NAM New App oach Me hodologies
PMF Po en ial o Mean Fo ce
QM Quan um Mechanics
QNAR Quan i a i e Nanos uc u e–Ac i i y Rela ionship
QSAR Quan i a i e S uc u e–Ac i i y Rela ionship
QSPR Quan i a i e S uc u e–P ope y Rela ionship
RNN Recu en Neu al Ne wo k
S(S) bD Sa e (and Sus ainable) by Design
SASA Sol en Accessible Su ace A ea
SCFT Sel -Consis en Field Theo y
VAE Va ia ional Au oencode
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
345
nanos uc u es wi h desi ed p ope ies (Sa e and Sus ainabili y-
by-design pe spec i e, SSbD) b ings along new oppo uni ies
bo h in he academic and indus ial con ex . None heless, signi -
ican challenges a ising om bo h he ex emely demanding
compu a ional models (physics-based and da a-based) and he
p ac ically unlimi ed numbe o possible combina ions o di e -
en subs ances and nano o ms ha lie ahead. E en o a single
ma e ial, he e is a ple ho a o possible bulk and su ace s uc-
u es, de ec s and e mina ions, each deli e ing unique p ope -
ies. This gene al challenge can only be ackled by an
in eg a ed app oach. He e, we c i ically analyze he p ocess asso-
cia ed wi h he in silico e alua ion o he ENMs desc ip o s, wi h
he aim o highligh ing challenges and oppo uni ies when using
physics-based modelling o gene a ing hem.
Ou discussion akes place unde he b oad pe spec i e o he
Eu opean and US legisla ion de elopmen and hei eme ging
s a egies ela ed o isk assessmen o nanoma e ials (e.g., he
EU Ma e ials Modelling Council, he EU REACH egula ion, US-
EU nanoEHS ini ia i e, he EU-US Nanoin o ma ics Roadmap
2030[8]). Despi e he well-accep ed ela ionship be ween physic-
ochemical p ope ies (i.e. desc ip o s) and he (eco)- oxicological
endpoin s, a comp ehensi e compu a ional desc ip ion o ENMs
and an unde s anding o he basic mechanisms behind hei
in e ac ion wi h biological media a e s ill e y challenging.
Fi s , we ocus on some o he (physics-based) compu a ional
app oaches o es ima ing ad anced desc ip o s a ime and
space scales ele an o nanosa e y. In his espec , we p o ide
an o e iew o he app oaches o in es iga ing ENMs elec onic
and a omis ic s uc u e ( hus ocusing on in insic ea u es) up o
he mesoscopic desc ip ion o in e ac ions wi h biological ma -
e , such as p o eins o cell memb anes, hence mo ing owa ds
mo e ex insic ea u es.
Second, mo e ecen da a-based models and app oaches o
nanosa e y assessmen a e analyzed. As illus a ed schema ically
in he op pa o Fig. 1, compu a ions can be pe o med a di e -
en space/ ime scales by sol ing app op ia e physical model
equa ions wi h ad anced desc ip o s collec ed om each simula-
o . Ad anced desc ip o s can be passed ac oss di e en ime and
space scales, ealizing a chain o mul iscale simula ions [9]. This
ision is ce ainly ascina ing, howe e , as discussed below, i
comes wi h o midable challenges: as opposed o he high-
accu acy da a ex ac ed by he physics-based models, he inhe -
en compu a ional cos is o en p ohibi i ely high. This leads
o a low da a a iance ha ende s he subsequen ansla ion
FIG. 1
TOP: Physics-based models ensu e high ideli y a a high compu a ional cos , making i possible o in es iga e only a es ic ed egion o he ENM design
space. Hence, he necessa y a iance is needed o desc ibe he la ge a ie y o pa ame e s o in e es (e.g., pa icle size, coa ing, e c.). Thus, p edic ing
ad anced desc ip o s o nanosa e y has o s em om di e en sou ces, such as li e a u e da a and c ude app oxima ion models cha ac e ized by a lowe
ideli y le el. BOTTOM: Disc epancy in ideli y and a iance is o be p ope ly o ches a ed, possibly using ML models [11].
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
346
o such da a in o inpu o QSAR/QSPR (o o he da a-based)
models ex emely di ficul i no impossible. De ails on obs acles
a e e y ele an scale a e discussed, and he cu en s a us and
challenges wi h he Eu opean egula o y con ex on ma e ials
modeling and in silico es ima e o desc ip o s o nanosa e y pu -
poses a e epo ed below [10].
Sa e and Sus ainabili y-by-design (SSbD) s a egy:
F om nano o m desc ip ion o sa e y modeling
Wi hin he Eu opean G een Deal, he Chemicals S a egy o Sus-
ainabili y (CSS)[12] iden ified se e al ac ions o educe nega i e
impac s on human heal h and he en i onmen associa ed wi h
chemicals, ma e ials, p oduc s, and se ices comme cialized o
in oduced on o he EU ma ke . In pa icula , he ambi ion o
he CSS is o phase ou he mos ha m ul subs ances and subs i-
u e, as a as possible, all o he subs ances o conce n and o he -
wise minimize hei use and ack hem. This objec i e equi es
no el app oaches o analyze and compa e all li e cycle s ages,
e ec s, eleases, and emissions o specific chemicals, ma e ials,
p oduc s, and se ices, and mo ing owa ds ze o pollu ion o
ai , wa e , soil, and bio a. The SSbD amewo k aims o suppo
he design and de elopmen o sa e and sus ainable chemicals
and ma e ials wi h esea ch and inno a ion (R&I) ac i i ies.
Al hough he sa e y o nanoma e ials has been o conce n o
he scien ific communi y o mo e han wo decades, he e is s ill
a limi ed numbe o alida ed and egula o y-accep ed al e na-
i e nano-specific app oaches o assessing hei human sa e y.
In ac , he cu en knowledge abou a ious ad e se e ec s
induced by nanoma e ials a e exposu e does no ye enable a
b oad de elopmen o he SSbD s a egy.
The sa e y and sus ainabili y assessmen
Reliable and e ficien me hods allowing, in a imely manne , o
assess exposu e isks should be fi s de eloped. In his con ex ,
he de elopmen o new al e na i e me hods combining
in i o, chemical analysis as well as in silico models o p edic
he po en ial ad e se impac o chemicals, including nanoma e-
ials, on human heal h is highly needed [13]. These ools a e
expec ed o be use ul o egula o s and policymake s; he e o e,
hey need o conside biologically plausible and egula o y-
ele an e en s essen ial o he occu ence o possible ad e se
ou comes [14]. In line wi h his poin o iew is he s a egy ha
in eg a es nanoin o ma ics models wi h he Ad e se Ou come
Pa hways (AOPs) [15,16]. The AOP is a amewo k ha desc ibes
a sequence o biological e en s ollowing s esso exposu e and
leading o a ious Ad e se Ou comes (AO). This concep links
Molecula Ini ia ing E en s (MIE) o he se ies o key cellula -
o issue-le el changes (so-called Key E en s, KE) ha culmina e
in he mani es a ion o AO. By suppo ing he iden ifica ion o
AOP- ele an e en s and in o ma ion ha migh be applied o
he weigh o e idence-based sa e y decisions, AOP can se e as
a amewo k o de eloping nanoin o ma ics models use ul o
egula o y ac ions. This would mean ha e en s ecognized as
c ucial o mani es nano-specific ad e se e ec s should be con-
side ed o modeling. As an example, ecen ly, Jagiello e al.
[15] p oposed a model ha linked he s uc u al p ope ies o
mul iwalled ca bon nano ubes wi h he ec ui men o p o-
inflamma o y media o s in o he lungs, ha finally leads o lung
fib osis acco ding o AOP173. Applying he AOP amewo k in
his model enables o unde s and e en s igge s ollowing nan-
o ube exposu e and occu ing in di e en biological o ganiza-
ions (molecula , cellula , issue, o gan, indi idual). To sum up,
wi h he de elopmen o he AOP-ancho ed nanoino o ma ics
models (models o key biologically plausible and egula o y-
ele an e en s), he usage and accep ance o hose app oaches
in making egula o y decisions abou he sa e y o nanoma e ials
exposu e can inc ease.
Addi ionally, he JRC (2022)[17] e iewed p e ious decision
amewo ks o ma e ials sa e y o in es iga e how sus ainabili y
(b oadly including social and economic aspec s) was conside ed
and o define a se o c i e ia o in eg a e sa e y and sus ainabil-
i y. Sus ainabili y assessmen can use con en ional sus ainabili y
me ics adap ed o nano echnology p oduc s and p ocesses. In
his con ex , S iebe o a e al. [18] used echno-economic and li e
cycle en i onmen al c i e ia o sus ainabili y assessmen , o
compa e al e na i e nanopa icle p oduc ion echnologies;
Ga cía-Quin e o and Palencia [19] analyzed con en ional quan-
i a i e sus ainabili y me ics and p oposed Li e Cycle Assess-
men (LCA) as a sui able me ic o compa e, op imize and
quan i y bio-based nanobio echnology and nanosyn hesis
p o ocols.
Regula o y ac ions o d i e SSbD chemicals and ma e ials
P edic i e in silico modeling, accessible and sea chable da abases
and quan i a i e ools o isk assessmen and p e en ion a e c i -
ical o he success ul implemen a ion o SbD/SSbD s a egies in
he nanosa e y con ex and o de elop ele an p o ocols, e e -
ence ma e ials, ealis ic in i o and compu a ional models, as
well as g ouping and ead-ac oss me hods. [20] Addi ional chal-
lenges s em om he need o assess he a e and eac i i y o nex
gene a ions o nanoma e ials [21] including sma nanoma e ials
and nano-enabled p oduc s [22]. New App oach Me hodologies
(NAMs), comp ise a wide ange o such models and ools [23]
o conduc obus and eliable chemical sa e y assessmen s and
educe animal es ing.
Wi h subs an ial unding, he Eu opean Commission has sup-
po ed NAMs o nanoma e ials, Re s. [24,25], acknowledging
hei po en ial in egula o y decision-making and inno a ion.
NAMs could inc ease egula o y p epa edness h ough fi - o -
pu pose da a and s anda dized es s specific o nanoma e ials
[26], and hei po en ial would be be e exploi ed wi hin ie ed
egula o y schemes. [27] Conside a ion o a ailable s anda ds (as
he OECD s anda d o QSARs [28] and ecommenda ions o he
Eu opean Ma e ials Modelling Council o alida ing in silico
ools is essen ial o p omo e hei adop ion o egula o y use.
On he echnical side, his also enables model in eg a ion in o
comple e compu a ional IATA wo kflows [29].
As p esen ed in Sec ion ‘Da a-based models o linking ENM
ea u es o sa e y’, he e is ich li e a u e on a a ie y o models
o he p edic ion o ENMs p ope ies and oxici y. The e e se
p oblem, i.e. he de elopmen o NM s uc u es ea u ing desi ed
o op imal p ope ies has also been conside ed in ew ecen
wo ks. As a ep esen a i e example, p elimina y wo ks consid-
e ed a condi ional deep con olu ional Gene a i e Ad e sa ial
Ne wo k (GAN) using compe i i e lea ning o sugges nanopho-
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
347
onic s uc u es wi h desi ed op ical p ope ies [30], and a GAN
o gene a e c ys alline po ous ma e ials, in pa icula pu e silica
zeoli e s uc u es [31]. Howe e , since he e is s ill much o do
o unde s anding he beha io o ENMs, se e al p ojec s ha e
been unded by he EU in he las yea s [32–35] o add ess sa e y
and isk go e nance issues.
Wi h his in mind, esea che s in NanoIn o maTIX [36] p o-
jec ha e de eloped an op imiza ion me hodology o guide he
sea ch o sa e ENMs and o suppo he applica ion o sa e( )-
by-design (SbD) app oaches [37]. The assessmen and sc eening
p ocesses use e ficien ep esen a ions coupled wi h quan i a i e
ools o simila i y assessmen o in es iga e mo phology-based
beha io [38]. The now implemen ed ool can be ained using
a ailable da ase s and can enable he assessmen o a ious
design op ions gene a ed du ing he op imal sea ch.
Haza ds o ENMs
As men ioned abo e, he physicochemical p ope ies o nanoma-
e ials may significan ly di e om hei bulk coun e pa s.
Despi e he beneficial echnological consequences associa ed
wi h his, ENMs migh exhibi haza dous e ec s on human
heal h o he en i onmen . The expe ience ga he ed om he
employmen o asbes os fibe s decades ago se es as a cau iona y
ale o he po en ial haza d o nanoma e ials. Asbes os we e
ex ensi ely used in a ious p oduc s wi h a wide ange o appli-
ca ions because o he ad an ageous p ope ies, like hea esis-
ance/isola ion and du abili y. Howe e , a e many yea s o
ubiqui ous quo idian p esence, he insidious ha m ul e ec s on
heal h became e iden . In his ega d, i was disco e ed ha a e
a long la ency pe iod asbes os we e able o cause lung cance and
meso helioma [39]. This e ec was linked wi h i s high-aspec
a io fibe s uc u e and us a ed phagocy osis. Ga he ing ma e-
ials by common ea u es leading o simila ad e se e ec s is a
fi s s ep o g oup and ead-ac oss. The e o e, in es iga ing he
po en ial haza ds o nanoma e ials is c ucial o gua an ee hei
sa e y enabling hei esponsible usage. Unde s anding hei
po en ial haza d is impo an o enginee ele an modifica ions
and unde ake he necessa y measu emen s o minimize expo-
su e, while s ill p ofi ing om hei a ious beneficial p ope ies.
Unde s anding hei po en ial haza d equi es desc ibing su ace
s uc u e, p ope ies and eac i i y, which define how hese engi-
nee ed nanoma e ials in e ac wi h each o he and wi h he en i-
onmen . The la e de e mines he end o agg ega e, which
a ec s a e and exposu e; mo eo e i shapes hei su ace p op-
e ies, hus al e ing in e ac ion among pa icles ( a e, exposu e)
and modi ying eac i i y. Al hough he e a e se e al exposu e
scena io o nanoma e ial oxici y, inhala ion is conside ed he
mos c i ical up ake ou e since i allows he pa icles o each
he sensi i e issues deep wi hin he lungs [40]. The e, he pa i-
cles could be aken up by lung cells o in e ac wi h he immune
sys em, leading o ha m ul e ec s. Asbes os is no he only 1-
dimensional ma e ial po en ially ha m ul. O he ma e ials, wi h
e y di e en chemis y and composi ion, sha e common ea-
u es. S udies on specific ypes o ca bon nano ubes sugges ed
ha hey may show simila oxici y as asbes os fibe s a lung
le el [41,42]. Simila ly, esea ch on sil e nanopa icles showed
geno oxici y and poin ed ou ha smalle -sized pa icles exhib-
i ed highe oxici y in in i o se ings [43]. I is impo an o
emphasize ha no all nanoma e ials a e associa ed wi h isks
o human heal h. Thei oxicological po en ial depends on di -
e en ac o s such as size, shape, unc ionaliza ion, besides
chemical composi ion. Ul ima ely, oxic e ec is a phenomenon
igge ed a he su ace o nanoma e ials, and how i in e ac s
wi h i s en i onmen . Bulk p ope ies do no ypically co ela e
wi h su ace p ope ies o s uc u e; he e o e, no di ec ansla-
ion o bulk cha ac e is ics can be made o su ace eac i i y.
Expe imen al cha ac e iza ion may be hampe ed by his, which
mus be aken in conside a ion when defining cha ac e iza ion
s a egies. F om a mo e me hodological pe spec i e, ma e ial
cha ac e iza ion should include de ailed in o ma ion on he su -
ace and i s de ec s, how he la e and he o me depend on he
unde lying bulk s uc u e and de ec s and how hose a e a ec ed
and, ul ima ely, shaped by he in e ac ion wi h he en i onmen
o biological media. The combina ion o hese p ope ies esul s
in coun less possible nanoma e ials ha can en e he ma ke ,
which can apidly o e whelm he cu en isk assessmen
p ocedu e.
Fo his eason, SSbD s a egies a e pa icula ly impo an , as
hey ha e he po en ial o acili a e and expedi e isk assessmen ,
necessa y o minimize he po en ial isks associa ed wi h he use
o nanoma e ials h oughou hei li ecycle. Mo eo e , SSbD
s a egies a e c ucial o ensu e he esponsible de elopmen o
nanoma e ials and hei applica ions.
The ole o ma e ials modeling in ENMs’haza d assessmen
I is wo h s essing ha accu a e physics-based modelling o
ma e ials is ypically no mean o di ec ly simula e he basic
mechanisms unde pinning ENMs oxici y. On he con a y, such
ools and me hods can be used o compu e p ope ies ha can
be e cap u e he complexi y o ENMs composi ion and he
influence o ex e nal condi ions on oxici y, as opposed o di fi-
cul , ime-consuming and expensi e expe imen s (when expe i-
men s a e possible). Despi e he benefi s o using ma e ials
modelling o calcula e nano-desc ip o s, e en un il a ew yea s
ago, he scale and complexi y o sys em simula ions we e chal-
lenging, and he e was a sho age o models o p edic impo an
p ope ies, such as he NM dissolu ion a e [8].
One may hus conclude ha i is highly desi able o base
nanosa e y assessmen upon accu a e in insic and ex insic
p ope ies, i.e. ea u es ha ep esen a pa icula s uc u e and
chemical na u e o he ENMs o in e es and how hese p ope -
ies a ec o a e a ec ed by hei (biological) en i onmen . In
he ollowing, such ea u es o ad anced desc ip o s and hei e al-
ua ion by means o physics-based models is discussed in mo e
de ail. In e pola ion- o ex apola ion-based me hods like QSAR
and ML will be se ed by a de e minis ic calcula ion o such
ad anced desc ip o s in a pa o he pa ame e space ha is
expensi e o ha d o assess expe imen ally, o when expe imen-
al da a a e unclea o incomple e. On he o he hand, physics-
based models may also p o ide di ec insigh in o he ela ion
be ween obse ed ensemble p ope ies (phenomena) and sys em
pa ame e s (simple and ad anced desc ip o s) ha s ems om a
sys ema ic compu a ional sc eening o one o mo e well-defined
design pa ame e s. Examples o phenomena conside ed in his
e iew a e p o ein abso p ion, NP agg ega ion and memb ane
binding. An ins ance o di ec insigh gained by compu a ional
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
348

means is he finding ha small nanopa icles (d, wi h d he
memb ane hickness, usually 4–5 nm) can simply pe mea e
h ough he memb ane, simila o small molecules, while la ge
nanopa icles (d) will be ully engul ed o w apped by he
memb ane upon binding. The binding cha ac e is ics o
nanopa icle sizes compa able o he memb ane hickness is s ill
unclea [44].
A sepa a ion o desc ip o s in o in insic and ex insic p o-
ides a s aigh o wa d basis o he selec ion o he mos app o-
p ia e physics-based model. Sec ion ‘Da a-based models o
linking ENM ea u es o sa e y’discusses he cu en knowledge
on ENMs and he (addi ional, ad anced) ea u es ha need o
be conside ed, as well as he p og ess on nanoma e ial ep esen-
a ions and da a-based models, like nano-QSARs, ha can ecei e
and p ocess he calcula ed nano-desc ip o s.
Da a-based models o linking ENM ea u es o sa e y
The c i ical ole o a gene al ep esen a ion o ENMs
In he a emp o gaining a deep a ionale om da a linking ENMs
and hei obse ed sa e y p ope ies, an essen ial p e equisi e is
he abili y o ep esen complex ma e ials in well s uc u ed and
machine- eadable o ma . Un o una ely, o da e, he lack o a
s anda dized seman ic cha ac e iza ion o he s uc u al ENMs
ea u es and en i onmen al a iables makes i di ficul o agg e-
ga e, cu a e and e alua e da a om di e en sou ces and o use
hem o simula ions o o aining new (da a-d i en) models;
hus making he meaning ul in eg a ion o da ase s a e y
demanding ask. Web da abases and eposi o ies o chemical
subs ances (e.g., ChEMBL, PDBe, ZINC15, Pubmed e c.) use s an-
da dized linea no a ions o subs ance iden ifica ion (SMILES,
SYBYL Line No a ion, o InChI) [45,46]. These no a ions a e also
employed in deep lea ning ools o guide he gene a ion o easi-
ble molecula s uc u es wi h desi able p ope ies (e.g., Gene a-
i e Ad e sa ial Ne wo ks (GANs), Va ia ional Au o-Encode s
(VAEs), Recu en Neu al Ne wo ks (RNNs), e c.). The ex ension
o hese no a ions o polyme s, mix u es, eac ions, e c. has also
been p oposed, bu ENMs en ail addi ional challenges compa ed
o con en ional chemicals. The key p ope ies o ENMs ha e a
s ong dependence on he physical and s uc u al ea u es. ENMs
cha ac e is ics such as he spa ial ela ionship be ween compo-
nen s, hei ela i e sizes, e c., all play an impo an ole in hei
inhe en and eme gen p ope ies. Wha is mo e, many o hese
p ope ies a e en i onmen -dependen , as hey a e a ec ed by
a ious ex e nal ac o s such as empe a u e and concen a ion
h ough non-linea dependencies, making ENMs desc ip ions
e en mo e sub le and complica ed.
A comple e ENM ep esen a ion, he e o e, would equi e he
inclusion o in o ma ion on many mo e aspec s han wha cu -
en linea no a ions p o ide. Ex ending he cu en ep esen a-
ions o include such ad anced in o ma ion con en is no a
s aigh o wa d ask. Lynch e al. [47] ha e ini ia ed an ex ensi e
discussion among a ious s akeholde s and p oposed a ame-
wo k o an InChI s anda d applied o ENMs as well as a oadmap
o i s de elopmen . They aimed o add ess he a ie y o com-
plex nanos uc u es, using a hie a chical app oach ha in o-
duces new laye s on he InChI no a ion o he size, shape,
c ys al s uc u e, and ligand binding o he ENM, and, possibly,
ex insic and su ace p ope ies. Recen ly, Blekos e al. p oposed
p inciples o a mo e accu a e, comple e, flexible and inc emen-
al ep esen a ion app oach. The p inciples we e demons a ed
h ough he de elopmen o ex ensions o nano-InChI o encode
mo phology/mix u e p ope ies and s a is ical dis ibu ions o
p ope ies and o s o e me ada a and enable hei euse.
The p oposed ep esen a ion amewo k could also p o ide
he heo e ical backg ound o g adually cap u e he eal pa icle
dynamics unde specific en i onmen al condi ions. [38] These
a e cu en ly unde conside a ion in he Nanoma e ials InChI
Wo king G oup (h ps://www.inchi- us .o g/nanoma e ials/)
and hei p oposal o a new InChI s anda d. A hope o he nea
u u e is ha an inc easing numbe o ad anced desc ip o s - as
hose discussed below in Sec ion ‘Physics-based models o
nanos uc u e cha ac e iza ion’- will be g adually inco po a ed
in o such s anda d no a ions.
Da a-based models o sa e y assessmen
In nanoin o ma ics, popula da a-based models include Quan i-
a i e S uc u e Ac i i y Rela ionship modeling (so-called nano-
QSAR/-QSPR) ha u ilizes Machine Lea ning (ML) and a ificial
in elligence o p edic he desi ed esponse (e.g., biological ac i -
i y, oxici y endpoin s o any physicochemical p ope y o in e -
es ). Those app oaches a e based on a se o compu a ional and/
o expe imen ally de eloped desc ip o s ep esen ing nanopa i-
cle s uc u e. As a ep esen a i e example, Wy zykowska, Mikola-
jczyk, e al. [48] p oposed a concep in which a nanos uc u e is
cha ac e ized by a iad ha desc ibes: (i) molecula s uc u e; (ii)
molecula desc ip o s; (iii) molecula p ope ies ha co espond
o chemical composi ion and chemical s uc u es o i s compo-
nen s. The p oposed iad co e s he cha ac e iza ion o he
in insic p ope ies o nanopa icle s uc u e (so-called sys em-
independen o in insic (nano-) desc ip o s (S-desc ip o s in
Fig. 2, le panel). [49].
Recen ad ancemen s in nano-QSAR modeling ha e in o-
duced hyb id models ha enhance p edic i e capabili ies by
in eg a ing mul iple modeling app oaches. These hyb id models
o en combine molecula dynamics simula ions da a wi h
machine lea ning echniques o be e unde s and and p edic
he complex beha io o nanoma e ials in biological sys ems
and hei in e ac ions wi h he en i onmen .[50,51] Al hough
QSAR/QSPR can be ce ainly ega ded as aluable ools o com-
plemen expe imen al s udies on chemicals and nanoma e ials,
hey come wi h mul iple challenges ha should be ca e ully ana-
lyzed. In pa icula , he e is an absence o comp ehensi e me h-
ods o cha ac e ize nanopa icles, which a e essen ial o a
s anda d QSAR p ocedu e o a dependable compu a ional
app oach ha accu a ely ep esen s he unique ea u es o
nanos uc u es –in sho , a lack o us wo hy nano-desc ip o s.
Fu he mo e, depending on he su ounding en i onmen ,
ENMs ea u es may change; hus, he da a-based models should
be de eloped based no only on he cha ac e iza ion o ENMs
chemical composi ion/chemical s uc u es o i s componen s
bu also on he influence o he en i onmen (including expe i-
men al condi ions, o biological medium). In o he wo ds, he
so-called sys em-dependen (ex insic) nano-desc ip o s o he
en i onmen (E-desc ip o s) should also be conside ed o
desc ibe nanos uc u e as a whole (Fig. 2, igh panel) [48,49].
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
349
In he ollowing wo subsec ions, we p o ide a b ie c i ical
e iew o in insic and ex insic desc ip o s in p epa a ion o
he mos ecen and ad anced compu a ional app oaches o es i-
ma e hem, as discussed in de ail in Sec ion ‘Physics-based mod-
els o nanos uc u e cha ac e iza ion’.
Sys em-independen in insic ea u es
Sys em-independen (in insic) nano-desc ip o s e e o he com-
posi ion, componen s’s uc u es and p ope ies ha may be mea-
su ed o calcula ed unde a well-defined, unchanging se o
condi ions. A fi s example o ad anced desc ip o s ha we e suc-
cess ully applied o da a-based modeling o nanopa icles was
de eloped based on quan um mechanical calcula ions (so-called
QM Desc ip o s) [52]. The QM Desc ip o s ha desc ibe he co e
chemis y o he s uc u e we e p oposed in 2011 by Puzyn
e al. [52] QM desc ip o s eflec he elec onic o m o a chemical
compound. They a e ob ained by applying quan um mechanics
o he app op ia e molecula model o an ENMs s uc u e. Du ing
he las 10 yea s, di e en esea ch g oups ha e been wo king o
de elop mo e sophis ica ed ypes o desc ip o s ha a e no
ela ed o de ailed a omis ic simula ions. Fo example, in 2012,
To opo e al. [53] p oposed SMILES-based op imal desc ip o s
based on he encoded one-, wo- and h ee-elemen SMILES a i-
bu es o a compound and can be calcula ed wi h he CORAL so -
wa e [54]. This idea was hen ex ended o he simplex
ep esen a ion o molecula s uc u e (SiRMS). He e, ano he ype
o desc ip o based on he Liquid D op Model (LDM desc ip o s)
was p oposed by Sizochenko e al. in 2014 [55]. The me hodology
o LDM desc ip o s calcula ion assumes ha an ENM can be ep-
esen ed as a sphe ical d op in which elemen a y molecules a e
igh ly packed. A he same ime, he densi y o clus e s is equal
o he pa icle mass densi y [56]. The p oposed me hodology is
based on he he modynamically mos s able uni cell o he con-
side ed c ys al s uc u e, ha is eplica ed in h ee dimensions.
A e wa ds, a sphe ical ENM is c ea ed by emo ing all a oms ou -
side he indica ed diame e . This is a clea simplifica ion: In ac ,
bulk-cu su aces will es uc u e, and e en a sphe ical pa icle
may exhibi egula egions as well as a a ie y o de ec s. [57].
Mo eo e , su ac an s may eo de he su ace di e en ly [58].
Clea ly, du ing he las decade, compu a ional scien is s and
nanoin o ma icians ac i e wi hin he Eu opean sa e y commu-
ni y ha e made a conside able e o o in eg a e knowledge om
exis ing EU comple ed and ongoing p ojec s wi hin EU FP7 and
HORIZON 2020 o de elop a mo e comp ehensi e app oach o
nanos uc u e cha ac e iza ion [59]. Howe e , we should also ec-
ognize ha se e al key challenges ela ed o he app op ia e ep-
esen a ion and desc ip ion o ENMs s uc u e a e s ill ahead.
Among o he s, hose aspec s ha e been highligh ed by esul s
p o ided by Mikolajczyk e al., Wy zykowska, Mikolajczyk, e al.
[60,49]. As a as in insic p ope ies o ENMs a e conce ned,
one o he main challenges is ela ed o he desc ip ion o he
complexi y o nanoma e ials composi ion, hus clea ly equi ing
he de elopmen and use o mo e ad anced compu a ional ools
capable o e alua e/es ima e desc ip o s ha a e di ficul o e en
impossible o access by cu en expe imen s.
While such aspec s a e discussed in de ail below wi hin
Sec ion ‘Physics-based models o nanos uc u e cha ac e iza-
ion’, an addi ional c i ical poin , pa ly ela ed o he la e com-
pu a ional assessmen o in insic ENMs ea u es, is ela ed o he
lack o de ailed ma e ial cha ac e iza ion in he published li e a-
u e. Too o en, indeed, only nominal composi ion, shape and
size o nanoma e ial a e epo ed, wi h hei possible coa ing
only aguely (i a all) defined. Clea ly, hose unce ain ies add
e en mo e complexi y and make benchma king o compu a-
ional ma e ial modeling pa icula ly di ficul .
Sys em-dependen ex insic ea u es
While using QSAR/QSPR o o he da a-based models, a second
challenge conce ns ep esen ing he influence o ex e nal condi-
ions (su ounding en i onmen ) [48]. A ecen ly published
FIG. 2
In insic and ex insic desc ip o s: In insic desc ip o s (i.e. sys em-independen , some imes also e e ed o as S-desc ip o s) e e o he composi ion,
componen s’s uc u es, and p ope ies ha may be measu ed o calcula ed unde a well-de ined, unchanging se o condi ions. On he o he hand, ex insic
desc ip o s (i.e. sys em-dependen , some imes also e e ed o as E-desc ip o s) ep esen he in luence o he su ounding en i onmen , and may be a ying
in ime. See ex and Sec ion ‘Physics-based models o nanos uc u e cha ac e iza ion’ o mo e de ails..
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
350
s udy [49] indica es ha he sys em-dependen (ex insic) nano-
desc ip o s, also e e ed o as en i onmen desc ip o s (E-
desc ip o s), a e much mo e c i ical o con olling and manag-
ing ou he p ope ies o ENMs han nanos uc u e cha ac e is-
ics hemsel es. Thus, in addi ion o s anda d cha ac e iza ion,
he expe imen alis should p o ide in o ma ion abou changes
in he s uc u e o he nanopa icles depending on he en i on-
men ( he su ounding condi ions). As a esul , nex o he co e
and coa ing, su ace p ope ies such as p o ein co ona o ma ion
(so-called “biomolecula co ona”) play an essen ial ole in cha -
ac e izing ENMs’beha io and may be conside ed i s finge p in
in a biological medium, see also Sec ion ‘Ex insic ad anced
desc ip o s: Mesoscopic le el’below o mo e de ails.
The sys em-dependen (ex insic) nano-desc ip o s a e c ucial
in desc ibing physicochemical p ope ies such as elec opho e ic
mobili y o ze a po en ial alue unde specified condi ions,
eflec ing he hyd ophobici y, biomolecula co ona, dissolu ion
a e, so p ion, su ace eac i i y, deg ee o agg ega ion/agglome -
a ion, o ENM pe sis ence.
De eloping he sys em-dependen (ex insic) nano-
desc ip o s is challenging because he nanos uc u e may change
du ing i s li e ime due o i s anspo h ough di e en en i on-
men s. In ac , nano-bio in e ac ions a e in p inciple d i en by
he nanoma e ials a e in biological and en i onmen al compa -
men s, meaning how ma e ials ansloca e, ac as ca ie s o u -
he oxican s and finally come in ouch wi h biological a ge s.
Unde his pe spec i e, he su ace cha ge and we abili y a e
conside ed he key de e minan s o he a e and beha io o
nanoma e ials dispe sed in he exposu e media.
Fu he mo e, he elec os a ic in e ac ions ha keep pa icles
dispe sed, p e en ing o p omo ing con ac wi h cell memb anes,
depend on he su ace po en ial shown a he slipping plane (Ze a
po en ial), as well as o he impo an p ope ies ha d i e nano-
bio eac i i y such as hyd ophilici y o he su ace in e ac ion
wi h biomolecules solubilized in he media. In his espec , he
iden ifica ion o he pH a which he Ze a po en ial is equal o
ze o (isoelec ic poin ) allows making hypo hesis on he ype o
acid/base beha io o su aces and on he p esence o cha ged
molecules specifically adso bed, as well as on he colloidal s abil-
i y o nano-dispe sed phases and o he occu ing o he e o-
agg ega ion phenomena [60–63]. In addi ion, he Ze a po en ial
is e y use ul o he design op imiza ion o su ace unc ionaliza-
ion s a egies applied o con ol nanopa icles eac i i y bo h o
nanosa e y and nanomedicine pu poses, because i is p edic i e o
he amoun and ype o coa ing ha masks su ace si es, p o id-
ing a new biological iden i y o he dispe sed phases [60,64].
Few s udies epo in silico models o he p edic ion o Ze a
po en ial based on physicochemical in insic p ope ies [60,65].
One o he mos p omising app oaches ha could cope wi h
he challenge o ex insic desc ip o s is an applica ion o a omis-
ic simula ions complemen ed wi h coa se-g ained models o
ENM and biomolecules (see he dedica ed sec ions below). In
such mul iscale app oach, he coa se-g ained models (nano and
mic oscale) a e pa ame e ized and possibly alida ed by he da a
ob ained om mo e de ailed models a smalle scales (a omis ic
and quan um chemical). In his con ex , no el desc ip o s can
also be de i ed om he s a is ics o adso bed molecules a e
an analysis o he biomolecula co ona [66,67].
Mode n machine-lea ning me hods
Machine lea ning (ML) is a b anch o a ificial in elligence (AI)
ha in ol es aining compu e p og ams o make p edic ions
o ake ac ions based on da a. In ML, algo i hms a e designed o
lea n om da a, ins ead o being explici ly p og ammed o pe -
o m specific asks. The idea is o p o ide he compu e wi h a la ge
amoun o da a, and hen use his da a o each he compu e how
o iden i y pa e ns, make p edic ions, o classi y new da a. A i s
co e, ML includes ou basic no ions: algo i hms, models, da a,
and aining. ML algo i hms a e designed o lea n om da a and
make p edic ions o ake ac ions based on ha da a. The algo-
i hms gene a e models, which a e ep esen a ions o he pa e ns
and ela ionships in he da a. The quali y o he model depends on
he quali y and quan i y o he da a used o ain i , as well as he
algo i hm’s abili y o lea n om ha da a. The e o e, he p ocess
o aining an ML model in ol es eeding i wi h labeled da a,
measu ing i s pe o mance, and efining he model un il i
achie es sa is ac o y accu acy on new, unseen da a.
The a ionale o using ML app oaches o nanosa e y assess-
men is based on he need o e ficien ly p ocess, analyze, and
ex ac meaning ul in o ma ion om possibly as amoun s o
da a gene a ed om bo h compu a ional and expe imen s. ML
algo i hms can lea n complex ela ionships and pa e ns om
hose da a se s, enabling esea che s o make p edic ions, op i-
mize ma e ial p ope ies, and iden i y no el ma e ials wi h
desi ed (e.g. less oxic) cha ac e is ics.
The ad an ages o using ML - as opposed o mo e adi ional
app oaches - include:
Accele a ed ma e ials disco e y and op imiza ion: ML algo-
i hms can quickly p ocess la ge amoun s o da a and iden i y
po en ial new ma e ial candida es o modifica ions o u he
in es iga ion.
Reduced expe imen al and compu a ional cos s: By p edic ing
ma e ial p ope ies and hidden pa e n iden ifica ion, ML can
help educe he numbe o expe imen s and simula ions
equi ed in he de elopmen p ocess.
Enhanced unde s anding o complex ma e ial sys ems: ML
can cap u e non-linea ela ionships and in ica e pa e ns
in da a, leading o be e insigh s in o he unde lying physics
and chemis y o ma e ials.
Howe e , he e a e also challenges and limi a ions associa ed
wi h such ML app oaches:
Da a quali y and a ailabili y: ML algo i hms ely on high-quali y
and abundan da a, which can be a limi ing ac o in ma e ials
science, whe e da a may be sca ce, noisy o he e ogeneous.
In e p e abili y and explainabili y: ML models can be com-
plex and di ficul o in e p e , making i challenging o unde -
s and he unde lying easons o hei p edic ions and build
us in hei ou comes.
O e fi ing and gene aliza ion: ML models may be p one o
o e fi ing, namely hey pe o m well on he aining da a
bu poo ly on unseen da a, hus educing p edic ing accu acy.
ML echniques ha e been applied o a wide ange o applica-
ions and fields including nanoin o ma ics [68]. In such a con-
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
351
ex , ML me hods ha e he po en ial o significan ly impac he
design, cha ac e iza ion, and sa e y assessmen o ENMs. How-
e e , he a ailabili y o high-quali y and abundan da a is s ill
an impo an aspec o add ess.
Among o he challenges, a pa icula ly impo an aspec asso-
cia ed o da a-based modelling in nanoin o ma ics is he p ope
handling o highly imbalanced da ase s. Imbalanced da ase s
a e cha ac e ized by a skewed class dis ibu ion (e.g., o e 1:100
obse a ions in he mino i y class compa ed o he majo i y
class), whe e usually he mino i y (unde ep esen ed) class is
he mos in e es ing one o p edic . Fo example, in an unbal-
anced da ase , he e could be a majo i y class o nanoma e ials
ha a e composed o a single elemen (e.g. gold nanopa icles)
and mino i y classes o nanoma e ials ha a e composed o mul-
iple o he elemen s. As a esul , when aining a ML o classi y
he nanoma e ials he model is likely o be biased owa ds he
majo i y class and may no pe o m well on he mino i y class
o high in e es .
To add ess he issue o class imbalance, a ious da a-le el and
algo i hm-le el app oaches ha e been p oposed o e he las dec-
ades [69–71]. Da a le el app oaches a e add essing class imbal-
ance ia esampling (unde sampling and o e sampling), as well
as ia e olu iona y algo i hms o sampling, ac i e lea ning o
selec ing he mos app op ia e da a poin s o mo e ecen ly
ad e sa ial lea ning algo i hms o new poin s gene a ion and
me a-lea ning [72]. Random unde sampling (i.e. emo al o
obse a ions om he majo i y class), Nea Miss, Tomek links
(i.e. emo al o bounda y obse a ions) a e examples o unde -
sampling (o downsampling) algo i hms. The disad an age o
he la e me hods is he loss o use ul in o ma ion as well as pos-
sible inc ease o he da a bias.
O e sampling app oaches include andom o e sampling
(mul iplica ion o obse a ions om he mino i y class), SMOTE
(finding nea es neighbo s o he mino i y class obse a ions and
adding poin s on he line joining he poin and he nea es
neighbo ) [73]. A e iew o SMOTE a ian s can be ound in
Re . [74] SMOTE ex ensions o handle mul iclass and mul ilabel
(MLSMOTE) classifica ion and eg ession (SMOTER) ha e been
p oposed [75], and, mo e ecen ly, DeepSMOTE [76] and G aphS-
MOTE [77]. O e sampling has he ad an age o e aining all he
in o ma ion and usually pe o ms be e han unde sampling.
Howe e , o e sampling may inc ease he p obabili y o
o e fi ing.
Ano he impo an aspec o be conside ed is ha accu acy is
no an app op ia e me ic o assessing model pe o mance on
imbalanced da ase s. Algo i hm-le el app oaches aim a modi y-
ing he loss me ic, assigning di e en cos s o penalize e o s in
each class (cos -sensi i e aining) o in oducing new algo i hms
which can inhe en ly deal wi h imbalanced da a. Ins ead o accu-
acy, he ecommended me ics a e con usion ma ix, p ecision
and ecall, F1 sco e, Kappa, A ea unde cu e (AUC) (see also
Re . [74]). Ensemble algo i hms using bagging and boos ing
(e.g., ee ensembles as Random Fo es ) a e known o be mo e
obus in imbalanced se ings. Recen li e a u e add esses han-
dling imbalanced da a (also known as long ail lea ning) by deep
lea ning me hods [78,79]. Imbalanced da ase s a e ypical in
high h oughpu sc eening [80–82] and chemogenomics [83].
ML algo i hms can be b oadly ca ego ized in o supe ised and
unsupe ised lea ning. We begin wi h a discussion o supe ised
lea ning in he ollowing subsec ion.
Supe ised lea ning
Supe ised lea ning wo ks on labelled da a, wi h he goal o
app oxima ing a unc ion ha maps inpu o labelled da a. A
p o o ypical example would be he abili y o link a se o ele an
in insic and ex insic ad anced desc ip o s o a numbe o
ENMs o hei obse ed (eco-) oxicological endpoin s, e.g. he
su i al and/o ep oduc ion a e o a chosen o ganisms. Supe -
ised lea ning echniques include a la ge a ie y o algo i hms
and me hods like Decision T ees, Random Fo es , Suppo Vec o
Machines, a ious classifie s like k-Nea es Neighbo s, Neu al
Ne wo ks and Ins ance-Based Lea ning me hods (see Fig. 3, le
panel) [84,85].
In one o he ea lies applica ions o ML me hods in manu ac-
u ed nanopa icles, Fou ches e al. used Suppo Vec o
Machine-based classifica ion and kNN-based eg ession o gene -
a e Quan i a i e Nanos uc u e–Ac i i y Rela ionship (QNAR)
models o p edic biological ac i i y p ofiles o no el nanoma e-
ials [86]. La e , Puzyn e al. [52] p esen ed a me hod o quickly
es he po en ial oxici y o enginee ed nanopa icles. They
applied a mul iple eg ession me hod combined wi h a Gene ic
Algo i hm (GA-MLR) o c ea e a model ha desc ibed he cy o-
oxici y o 17 di e en ypes o me al oxide nanopa icles o bac-
e ia Esche ichia coli [52]. Ge nand and Casman pe o med a
eg ession- ee-based me a-analysis on oden pulmona y oxic-
i y exposed o uncoa ed, non- unc ionalized ca bon nano ubes.
They epo ed he applica ion o Reg ession T ee models, Ran-
dom Fo es models, and a andom- o es -based dose– esponse
model [87]. Winkle e al. used no el spa se ML me hods o
model he biological e ec s o nanopa icles wi h a ious com-
posi ions, including i on oxide nanopa icles and gold nanopa -
icles. They employed Bayesian neu al ne wo ks using bo h
linea and nonlinea ML me hods [88].
E olu iona y app oaches ha e also been explo ed. Le and
Winkle [89], o example, e iewed he use o a ificial e olu-
iona y me hods o he iden ifica ion and op imiza ion o no el
ma e ials. They epo uses o gene ic algo i hms o in es iga e
he p ope ies o bime allic co e–shell and i anium dioxide
nanopa icles [90,91]. Ma inez e al. p esen ed decision ee
models based on e olu iona y algo i hms ha classified
nanopa icle agg ega es in o mo phological classes [92]. kNN
algo i hms ha e also been used o model he oxicological p op-
e ies o nanoma e ials. Wang e al. used a kNN algo i hm o
de elop QNAR models o biological ac i i y p ofiles like cellula
up ake in a ious human cells and he abili y o induce oxida i e
s ess [93]. Ko alishyn e al. used kNN, andom o es , and neu al
ne wo k me hods o gene a e models o he analysis o eco/ ox-
icological and physicochemical p ope ies o me al and me al
oxide nanopa icles [94].
Va ious Neu al Ne wo k a chi ec u es ha e also been in es i-
ga ed in ela ion o p edic i e nanoin o ma ics. Gomez-
Bomba elli e al. ained a Deep Neu al Ne wo k o au oma ically
gene a e no el chemical s uc u es and demons a ed hei
me hod on small d ug-like molecules [95]. Ha aminia e al. used
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
352
e ized by he nanopa icle’s diame e and in e ac ing wi h o he
beads acco ding o analy ical equa ions desc ibing he in e ac ion
po en ial (e.g. fi ed o calcula ed exp ession by means o classical
Molecula Dynamics o a nanopa icle pai , see Fig. 4, panel C).
Applying his coa se-g aining p ocedu e makes i possible o
de e mine new molecula desc ip o s uling pa icle agg ega ion
ha could be ed in o QSAR models [59]. Addi ional challenges
aced by de eloping gene al compu a ional and heo e ical mod-
elling o unde s anding he nanosa e y o ENMs s em om he
ole o he en i onmen in de e mining he obse ed oxici y.
Once he ENM en e s a li ing o ganism, i ge s in con ac wi h
he biological molecules, in pa icula p o eins and lipids.
Nanopa icle-p o eins and nanopa icle-lipids in e ac ion
could in heo y be in es iga ed using B ownian Dynamics simula-
ions, which can un e en on a 16 co e wo ks a ion, p o ided ha
all pai in e ac ions a e known; in p ac ice, his has no ye been
done because he calcula ion o he ee ene gy p ofiles o each
pai by all-a om MD is oo compu a ionally demanding. Anyway,
his could be an in e es ing a emp o b idge he gap be ween he
molecula simula ions scale and he expe imen al scale.
The oldes app oach o compu a ionally de e mining ma e ial
p ope ies, i.e. he con inuum mechanics pionee ed in he 19 h
cen u y by Cauchy, is in ac mos sui ed o sc eening pu poses,
since i combines modes compu a ional cos s o ealis ic sys em
sizes wi h a ew e ec i e sc eening pa ame e s. This sc eening idea
is a he basis o con inuum Sel -Consis en Field Theo y (SCFT),
which was de eloped o desc ibe phase beha io and phase sepa-
a ion dynamics in block copolyme s ( ep esen ed as flexible
chains) based on an implici molecula ep esen a ion [133].
Rigid objec s like nanopa icles ha e also been inco po a ed
in o SCFT, p ima ily o he pu pose o modelling polyme
nanocomposi es [134], and we e e o ea ly pape s o de ails
abou he di e en app oaches [135–137]. While hese field-
based me hods possess a clea ad an age o e ficiency o e
pa icle-based me hods like AAMD and CGMD, which s ems
om he choice o deal wi h ensembles a he han indi idual
chains, and SCFT in e ac ions a e o he desi ed many-body ype
by defini ion, his is o se by he se ious disad an age o no
being able o ep esen specific in e ac ions a he molecula le el
and ha ing no access o con o ma ional de ail. In addi ion, he
commonly used excluded olume in e ac ions in SCFT do no
allow o phase ansi ions ha can play a ole in memb ane
binding p ocesses. Only ecen ly, hyb id pa icle-field (hPF)
app oaches such as hPF-MD [138] and single chain in mean field
(SCMF)[139] ha e in oduced he abili y o combine pa icle-
based (a omis ic o segmen al) molecula de ail wi h he e fi-
ciency and mul i-body na u e o Hamil onians om con inuum
heo y like SCFT. Phase ansi ions and/o coexis ence ha e also
been added ecen ly [140]. Un il hese hyb id me hods o e a
alida ed solu ion o he need o combine e ficiency wi h speci-
fici y and molecula de ail, we conclude ha SCFT is use ul o
he in es iga ion o gene al phenomena, bu no sui ed o he
ex ac ion o ex insic ma e ial desc ip o s.
A popula and e y e ficien ep esen a ion o lipid memb anes
is ha o a hin elas ic shee wi hou any molecula de ail. Shape,
dynamics and esponses o de o ma ion a e dic a ed by a con in-
uum Hel ich ee ene gy ha depends on a ew collec i e p ope -
ies like bending igidi y and la e al memb ane ension, which
can be di ec ly ela ed o specific memb ane composi ions ia
pa icle-based simula ion. An ex ended Hel ich model de eloped
la e p o ided s aigh o wa d condi ions o nanopa icle up ake,
i.e. he balance o ene gy needed o s e ch and bend he mem-
b ane a ound he nanopa icle and he ene ge ic gain o nanopa -
icle binding. The la e is p o ided by he adhesion ene gy
densi y o (coa ed) nanopa icles in he con ac egion and, as
was la e ound, also in nea -con ac egions. I should be no ed
ha ex ac ing adhesion ene gy densi y o eal ma e ials om
mo e de ailed desc ip ions is, un o una ely, a a om simple
ask [141]. Ea ly on, Dese no e al. used he con inuum model
o show ha a ensionless memb ane can only adop wo s a es:
an unbound s a e whe e he memb ane is fla , o a s a e whe e
he nanopa icle is ully w apped by he memb ane [142,143].
Accoun ing o ac o s ha we e missing in his o iginal
Hel ich-based analysis, such as he memb ane hickness and
in e ac ion ange, Raa z e al. and Spangle e al. ound ha also
pa ially w apped cases could be s able [144,145]. While p o id-
ing impo an ene ge ic insigh , and, he e o e, being o po en ial
use o es ic ed bu quick sc eening, he lack o lipid de ail
al eady se iously hampe s he use o such me hods o ex ac ing
ex insic ma e ial desc ip o s. Ano he disad an age is ha he
na u e o he memb ane de o ma ion upon nanopa icle binding
is no an ou come bu equi ed a p io i, in oducing a isk o o e -
looking al e na i e binding mechanisms. The his o ical solu ion
o his issue is o employ highly coa se-g ained pa icle-based
models wi h implici sol en o s udying gene ic memb ane
dynamics and nanopa icle-memb ane in e ac ions. Using such
a model, he w apping cha ac e is ics o nanopa icles up o
40 nm was conside ed, i.e. he en i e ange o he mechanism
is expec ed o swi ch, and a discon inuous ansi ion om pa ial
o ull w apping was p edic ed a ound 10 15 nm nanopa icles
[145,146]. Since hese models lack sol en , which is known o
modula e he ( ee) ene gy landscape, and hey also lack he eso-
lu ion o dis inguish be ween di e en lipids, hey do no ep e-
sen a decisi e s ep o wa d in he sea ch o accu a e desc ip o s.
Summa izing, one may conclude ha hese e ficien molecule-
based mesoscopic me hods a e use ul o ex ac ing in o ma ion
abou gene al balances and mechanisms o sys ems o ele an
size, bu hey we e ne e mean o p o ide ex insic nanopa icle
desc ip o s. Jus his, balancing (chemical) in o ma ion and e fi-
ciency wi h he aim o e ain he necessa y de ail, is he main pu -
pose o ecen de elopemen in sys ema ic coa se-g ained
me hodology. Al hough he e a e se e al ways o pe o m sys em-
a ic coa se g aining, depending on he cha ac e is ics o he e e -
ence a omis ic sys em ha one wan s o ep oduce, hey a e all
based on lumping g oups o a oms in o CG pa icles o beads.
The mos popula me hod, CG Ma ini, combines 2–4hea y
a oms in a single bead. Gene a ing a desc ip ion in e ms o CG
beads does no only educe he compu a ional load, bu i also
so ens he in e ac ions. As a esul , also he sys em e olu ion is
significan ly accele a ed. Me hods based on such ypes o coa se
g aining ha e been applied o s udy lipid pa i ioning in gene al,
he binding o elas ic nanoshells [147,148], he adhesion o aniso-
opic nanopa icles [149,150] and o unc ionalized nanopa icles
[151,152]. Also he ole o bending and adhesion in he dis ibu-
ion o mul iple nanopa icles inside he memb ane has been
in es iga ed, bo h in e ms o a gene ic ep esen a ion in a highly
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
359

CG me hod [153] and o mo e chemically esol ed CG ep esen-
a ions [154–156]. Ye , while hese CGMD s udies ha e ei he been
designed o p ope ly ep esen a pa icula expe imen al nanopa -
icle o o ob ain insigh in o mo e gene al binding mechanisms,
e y ew ha e ocused on he challenge o de eloping a ans e -
able ep esen a ion o map om he a omis ic o he coa se-
g ained domain. Ye , de e mining such a map ha is alid o all
nanopa icles o he same ma e ial is a p e equisi e o he ex ac-
ion o ad anced desc ip o s and ends ha enable ex apola ion.
One e y ecen example o such a new de elopmen is he
special CG nanopa icle ep esen a ion wi hin he amilia Ma -
ini CG app oach ha is equi ed o s udying binding and
ansloca ion pa hways o ealis ic sil e nanopa icles ac oss sol-
a ed lipid ba ie s in he lungs, see Fig. 5. Whe eas he mod-
elling communi y has hus a gene ally app oached he
undamen als o such la ge-scale phenomena ia implici -
sol en con inuum [142,143] o highly coa se-g ained desc ip-
ions [145,146,157], a co e–shell CG ep esen a ion was de el-
oped ha is ans e able wi h espec o size and enables he
simula ion o ele an nanopa icle sizes including sol a ion
e ec s, in he size ange whe e in e es ing swi ching in binding
beha io is expec ed [158]. The de elopmen o his ans e able
map was based on ma ching po en ials o mean o ce (PMFs) o
sil e nanopa icles ob ained using all-a om molecula dynamics
[120]. The sys ema ic de elopmen o ans e able CG ep esen-
a ions o he o he ma e ials, such as silica and i anium diox-
ide ha we e also conside ed in a omis ic s udies, is a u u e
desi e. As he de e mina ion o binding ee ene gies o CGMD
NP by s anda d me hods like umb ella sampling ge s p ohibi i e
wi h inc easing size, s ing me hods could be employed as an
al e na i e [159].
A key equi emen in nanosa e y assessmen is how o include
sys ema ically a de ailed molecula desc ip ion o ENM-p o ein
in e ac ions. In biological en i onmen s, p o eins o ganize on
ENM su aces o ming he so-called nanopa icle p o ein co ona
(NPC) s uc u es which play a cen al ole in biological in e ac-
ions and nano oxici y [160–162]. The NPC o ma ion a ound
a a ie y o nanopa icles was e idenced and cha ac e ized in
e ms o i s biochemical composi ion by se e al expe imen al
s udies [163,164]. Howe e , i s e ec s on biological in e ac ions
and implica ions o nanosa e y conside a ions emain la gely
unknown [163,164–169]. Simila ly as abo e, a main esea ch
challenge is how o de elop e ficien ye accu a e compu a ional
me hods and ools ha can b idge he gap be ween a de ailed,
molecula -le el desc ip ion o ENMs in e ac ing wi h sol en s
and biomolecules such as p o eins a an a omis ic le el, and
he much la ge scale (i.e., ens o hund eds o nanome e s o
mic ome e s) co esponding o biological s uc u es such as
NPCs o cellula memb anes [170]. Coa se-g ained compu a-
ional me hods o p o ein-co e ed nanopa icles a e o en lim-
i ed o modeling en i e p o eins as single pa icles. Such
models a e success ul in showing how nanoma e ials ype, size,
and shape can lead o di e se p o ein composi ion o he NPC
[171,172]. Howe e , de ailed a omis ic aspec s o modeling p o-
ein in e ac ions a e equi ed o calcula e o he key
expe imen ally- ele an mesoscopic desc ip o s such as he
FIG. 5
(a) All-a om MD o nanopa icle/memb ane/wa e sys em, (b) all-a om MD binding ee ene gy o h ee di e en sizes o Ag nanopa icles, (c) all-a om MD
binding ee ene gy o silica o h ee nanopa icle sizes (g aphs ep oduced om Re . [120] wi h pe mission om he Royal Socie y o Chemis y). (d)
S anda d uni o m CG model and he new co e–shell CG model, ep oduced wi h pe mission om Singhal e al., MDPI Nanoma e ials, 2022[158]. (e) Wi h
inc easing hyd ophobici y, he mechanism o di ec inse ion in o he model lung memb ane swi ches o a w apping mechanism..
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
360
hyd ophobic ac ion o he sol en accessible su ace a ea
(SASA
H
) (see Fig. 6(a-b-c)). Recen de elopmen s in a omis ic
and mul iscale compu a ional me hods allow unique oppo uni-
ies o p obe he de ailed molecula mechanisms ha modula e
in e ac ions a bio-nano in e aces [173,174,29]. This app oach
can be ex ended o he calcula ion o mesoscopic biophysical
desc ip o s o an NPC and elies on simplified models allowing
u he compu a ional s udies o p o ein in e ac ions wi h ENMs
(see Fig. 6(a-b-c)) [163]. This has he po en ial o (i) un eiling he
ole o specific p o eins in NPC’s s abili y and biophysical p op-
e ies (e.g., hyd ophobic su ace a ea, cha ged pa ches), and (ii)
quan i ying he way in which nanopa icle co ona p ope ies
and p o ein–p o ein in e ac ions in he co ona a e modula ed
o di e en nanopa icle ypes. Fi s , all-a om molecula dynam-
ics (MD) simula ions o key plasma p o eins (e.g., human se um
albumin, fib inogen, immunoglobulin gamma-1 chain-C, com-
plemen C3, and apolipop o ein A1) can be used o s udy adso p-
ion on ypical nanopa icle su aces (e.g., i anium dioxide o
silica). Fo bina y p o ein–p o ein in e ac ions (e.g., only wo
in e ac ing p o eins) i is possible o pe o m exhaus i e a omis-
ic MD simula ions, bo h in he icini y o nanopa icle su aces
and in bulk o compa e di ec ly he esul s and in e he influ-
ence o he p esence o specific nanopa icles on he dynamic
and he modynamics aspec s o p o ein–p o ein in e ac ions.
Fig. 6(c)) illus a es he possibili y o iden i y esidues c ucial o
p o ein-nanopa icle in e ac ions in a specific sys em (he e,
human se um albumin- i anium dioxide). In he second s age,
he molecula mechanisms o p o ein-nanopa icle in e ac ions
a e p obed by looking a he dynamic and s uc u al p o eins
o se e al p o eins (and possibly lipids) in he c owded en i on-
men o nanopa icle co onas, using also molecula docking sim-
ula ions and, depending on sys ems size, coa se-g ained
simula ions o mix u es o mul iple p o eins [175,176] ha can
in es iga e he o ma ion o he p o ein laye on he nanopa i-
cle su ace, as illus a ed in Fig. 6(d-e- ). P elimina y s udies on
mul i-p o ein docking on nanopa icles, sugges ha knowledge
o p o ein composi ion and con o ma ions (e.g., efined om
MD simula ions) can be used o es ima e he o e all biophysical
p ope ies o NPCs, such as he hyd ophobic ac ion o hei
sol en -accessible su ace a ea, and su ace cha ge dis ibu ions
[177]. Ou s anding challenges in modeling he in e ac ions o
biological molecules in con ac wi h ENMs a e o ex end he
capabili y o docking p og ams o include mo e han jus a ew
ens o p o eins [178–181], and o include de ailed in o ma ion
on he specific co ona molecula composi ion ha is seldom
a ailable [170]. Addi ionally, besides p o eins, i is expec ed ha
FIG. 6
(a) Schema ic a omis ic model o an NPC. A ”so co ona”laye (dashed line) o loosely bound p o eins su ounds a ”ha d”co ona laye (con inuous black line),
p oximal o he ENM’s su ace. E en o sphe ical nanopa icles, he o e all shape and biophysical p ope ies o he NPC su ace will depend on i s
composi ion. (b) Compa ing alues o he SASA
H
o a ious co onas a ound a 4 nm sphe ical silica nanopa icle (dashed) wi h he alues calcula ed o simila
p o ein agg ega es wi hou including a nanopa icle (con inuous). (c) A omis ic co ona models allow he iden i ica ion o p o ein esidues ha may play
signi ican oles a he nanopa icle-p o ein in e aces. (d-e- ) Building an a omis ic model o an NPC by sequen ial docking o p o ein s uc u es (mucin, p e-
equilib a ed using MD) on a sphe ical silica nanopa icle. Mesoscopic desc ip o s such as SASA
H
can be es ima ed as s a is ical a e ages o e esul s om
docking mul iple ep esen a i e s uc u es.
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
361
u u e s udies will also include (i) o he componen s such as
lipids [182] and glycans [183] which play pi o al oles in ENMs
up ake and could also be key o he modeled ENM sys ems, as
well as (ii) an accu a e desc ip ion o he co esponding su ace
unc ionaliza ion [184]. Finally, me allic nanopa icles dese e
a special men ion as hey a e ou inely employed in cance he -
apy, whe e hey need o be selec i ely deli e ed o he umo is-
sues. Among all me als, gold nanopa icles a e widely used as
adiosensi izing agen s because o hei biocompa ibili y and
simplici y o syn hesis [185]. Coa ed me allic nanopa icles a e
used in ca alysis, sel -assembly, imaging, d ug deli e y, and sens-
ing applica ions. Me allic nanopa icles a e e y sensi i e o he
local en i onmen because o a phenomenon called “localized
su ace plasmon esonance”de i ing om he collec i e oscilla-
ion o su ace elec ons. [186] When coa ed wi h a monolaye
ligand, he me allic nanopa icles’p ope ies such as me al
educ ion and colloidal s abili y can be adjus ed o he desi ed
applica ion [187]. A omis ic and mesoscale simula ions allowed
an unde s anding o he a ypical dis ibu ion o mul iple ligands
on gold and sil e nanopa icles obse ed in he expe imen
[188], as well as he adso p ion o biomolecules on gold nanopa -
icles o di e en sizes [189]. We epo in Fig. 7 an example o
how molecula modelling simula ions (he e all-a om molecula
dynamics simula ions) can be used o in es iga e he beha io
o me allic nanopa icles, in pa icula , o cap u e adso p ion
phenomena o polyme s. In e es ingly in [190], i is clea ly
shown ha he design o shape and opology o he su ace in
me allic nanopa icles is a a he e ec i e s a egy o con ol
he p e e en ial polyme coa ing in some pa icle egions as com-
pa ed o o he s. Fu he mo e, using simila simula ions ools,
o he wo ks [188] ha e in es iga ed he p ecise pa e ning o
coadso bed su ac an s on sil e and gold nanopa icles. Hence,
by a ge ing mo e e ec i e con ol on he c ys allog aphic ea-
u es o me allic nanopa icles, we can en ision an imp o ed
con ol o ENMs coa ing hus also significan ly a ec ing hei
oxicological p ope ies.
Discussion: Challenges and pe spec i es
On one hand, om he abo e o e iew, i clea ly eme ges ha
he accu acy in p edic ing possible haza ds o ENMs c i ically
elies upon he abili y o use ea u es beyond wha can be
accessed in ypical expe imen al es s and cha ac e iza ion.
Impo an ly, hose ea u es depend on in insic and ex insic
p ope ies aiming a desc ibing bo h ma e ials and biological
en i onmen s.
None heless, when i comes o he di ec link o ENMs o hei
expec ed oxici y endpoin s, i is ai o say ha he cu en s a-
us o he de elopmen o he ha dwa e and algo i hms does no
allow a b u e- o ce assessmen o nanosa e y. A mo e iable
app oach is expec ed o be he calcula ion o in insic and ex in-
sic ad anced desc ip o s using a ple ho a o me hodologies
mos ly de eloped and used in o he fields (e.g. ma e ials model-
ing and biochemis y) o ex ac inpu ea u es o da a-based o
s a is ical models (e.g. QSAR, ML algo i hms) o finally link hem
o he oxici y endpoin s.
FIG. 7
Uppe panel: MD snapsho s o sel -assembly simula ions o 60 PLGAs and he Au nanopa icle in aqueous solu ion a 0, 0.5, 50, and 100 ns, ep oduced wi h
pe mission om Cappabianca e al., ACS Omega, 2022[190]; Bo om panel: Equilib ium s uc u es ob ained by mesoscale simula ions o sel -assembly o
bina y mix u es o su ac an s wi h a ying leng h di e ence o bulkiness di e ence on a sphe ical nanopa icle. Da k ( ed) beads and ligh (yellow) beads
ep esen head g oups o he wo species o su ac an s, om Re . [188]. See ex o mo e de ails..
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
362
Howe e , e en he la e app oach comes wi h o midable
challenges, mos ly associa ed wi h he size and complexi y o
ENMs o p ac ical use. On one hand, ENMs o expe imen al
in e es may ha e dimensions o de s o magni ude la ge han
hei compu a ionally a o dable coun e pa s. On he o he
hand, p ecise composi ions o pa icles and hei coa ing a e
o en known wi h li le de ail le el: This calls o a c ucial e o
o he ele an scien ific communi y in u u e expe imen al
wo ks whe e, in addi ion o he aluable measu emen o oxico-
logical endpoin s, a mo e comp ehensi e cha ac e iza ion
beyond nominal alues o ENMs is eques ed.
Fu he mo e, cu en ly, an in e es ing (and pe haps necessa y)
app oach seems o be he hyb idiza ion o pu e physics-based mod-
els wi h dispa a e da a sou ces. In pa icula , due o a p ac ically
unlimi ed numbe o di e en ENMs wi h g ea chemical and geo-
me ical a ie y, a uly ex ended adop ion o ad anced desc ip o s
in da a-based models o nanoin o ma ics looks inconcei able
wi hou le e aging high-fideli y mul iscale modeling da a om
bo h li e a u e and c ude o analy ical (ye compu a ionally e fi-
cien ) app oxima ions models. As a ep esen a i e example, biased
classical molecula dynamics simula ions can ce ainly be used o
accu a ely compu e he Po en ial o Mean Fo ce be ween nanopa -
icle pai s. A he same ime, classical app oaches such as he heo y
o De jaguin-Landau-Ve wey-O e beek (DLVO) canno be dis-
ca ded and e o s should be de o ed o finding new app oaches
capable o o ches a ing and in eg a ing such mul i-fideli y and
mul i-sou ce da a. Ano he example o a simila syne gy has been
desc ibed abo e in he manusc ip and i has o do wi h CGMD
based simula ions o cell memb anes and he co esponding con-
inuum Hel ich ee ene gy models.
One possibili y would be he adop ion o desc ip o s om
c ude app oxima ion models, cha ac e ized by a lowe fideli y
le el and a ailable li e a u e da a as a subse o ea u es o ML
models whe e physics-based model esul s a e used as aining
se s. Such an app oach has been p o en success ul in signifi-
can ly imp o ing ML model p edic ions in he p esence o a
small o incomple e aining da ase [11,193]. We hus en ision
a mo e comp ehensi e and mul i-laye ed app oach whe e
de ailed physics-based models (fi s laye ), li e a u e/expe imen
da a (second laye ), and c ude app oxima ion models ( hi d
laye ) a e syne ge ically managed o he es ima e o ele an
desc ip o s by means o su oga e (s a is ical) and ML models
(see Fig. 1, bo om panel). In his espec , u he esea ch is
eques ed o in es iga e o wha ex en he small a iance o
high-fideli y da a om physics-based model p edic ions com-
bined wi h he high- a iance (and low-fideli y) o o he sou ces
can deli e desc ip o s p edic ions wi h high/medium fideli y
and a iance ha a e sui able o QSAR/QSPR models.
I is also wo h s essing ha , beyond he me e use ulness o
ad anced desc ip o s o da a-based models a ge ing nanosa e y
assessmen , u he imp o ed physics-based models desc ibing
phenomena om he elec onic up o he mesoscopic le el can
o e he unique oppo uni y o gaining insigh s a he nano-
le el, which a e ha dly accessible by expe imen al echniques
and he e o e emain c i ical o i) un eiling basic mechanisms
behind he possible haza dous cha ac e o ENMs; ii) possibly
complemen in o ma ion om less de ailed and less demanding
models.
Fo all he abo e easons, we expec and hope ha he p o-
g ess o high-pe o mance compu ing powe combined wi h
ad anced ools o accele a ion o a omis ic simula ions [194]
can soon calcula e ad anced desc ip o s mo e compe i i ely as
compa ed o expe imen s in e ms o bo h he amoun and qual-
i y o gene a ed da a.
Ano he in e es ing aspec o highligh is ela ed o he possi-
ble exploi a ion o he la ge body o knowledge a ailable in he
con ex o mode n compu a ional app oaches o in es iga e
nanopa icles in o he echnological fields such as ca alysis and
ma e ials science [195–196,77,197–200]. Specifically, we e e o:
1. High- h oughpu sc eening o nanopa icles in d ug deli e y
[201]
2. Ba coded nanopa icles o high h oughpu in i o disco e y
o a ge ed he apeu ics [202].
3. Compu a ional high- h oughpu sc eening o alloy nanoclus-
e s o elec oca aly ic hyd ogen e olu ion [203]
I would be desi able ha he Sa e and Sus ainabili y-by-
design s a egies in oxicology could ake ad an age o his
knowledge and ools, combining e ficien ly biology and medi-
cine wi h physics and chemis y.
Fu he mo e, an e en la ge amoun o da a on nanosa e y o
ENMs will likely be gene a ed in he nea u u e. I is he e o e o
inc easing impo ance o comply wi h he FAIR p inciples, so
ha me ada a and da a can be e ficien ly eused in da a-based
models o p edic ing he haza d o ENMs. Specifically, in his
espec , he ole o da abases o collec ing and s o ing a la ge
amoun o cu a ed and well s uc u ed da a is likely o play a
majo ole soon. As such, we discuss one p o o ypical case s udy
in he ollowing subsec ion.
Finally, as a long- e m goal, we expec ha nanoin o ma ics
models based on ad anced desc ip o s could be in eg a ed wi h
AOPs, o be e assess he po en ial exposu e h oughou he
en i e li e o he nanopa icles. In his espec , mode n g ouping
s a egies aking in o accoun he mode o ac ions and de eloped
based on ML echniques p ocessing da a om omics s udies
appea pa icula ly p omising and he e o e i will be specifically
discussed below in a dedica ed subsec ion.
Da abases: The eNanoMappe case s udy
S o age and o ganiza ion as well as he o ganiza ion o cu a ed
da a om dispa a e sou ces wi hin da abases appea c i ical o
nanosa e y assessmen . To his end, below we e iew a p o o yp-
ical case s udy. The eNanoMappe da abase is an open-sou ce
chemical subs ance da a managemen solu ion [204], adop ed
by mo e han 20 Eu opean p ojec s and acili a ing he Findable,
Accessible, In e ope able and Reusable (FAIR) da a collec ion and
euse o he nanosa e y communi y. To p o ide agg ega ed find-
abili y, accessibili y, and in e ope abili y ac oss p ojec -specific
da abases, he Nanosa e y Da a In e ace (h ps://sea ch.da a.
enanomappe .ne ) was c ea ed, and cu en ly ep esen s one o
he la ges sea chable nanosa e y da a collec ions [205]. The
eNanoMappe is based on da a and so wa e o iginally de eloped
o ep esen indus ial chemicals and ela ed expe imen al o cal-
cula ed da a. I was one o he fi s chemin o ma ics pla o ms o
o e open REST Applica ion P og amming In e ace (API) sup-
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
363
po ing in eg a ed se ices such as da a, desc ip o calcula ions,
and ML [206,207]. A isual ep esen a ion o he eNanoMappe
da a model is epo ed in Fig. 8, whe e subs ances a e cha ac e -
ized by names and IDs, which can be mul iple, he composi ion
e e s o he componen s o he ma e ial (co e, coa ing, chemical
s uc u e), each o hem ha ing di e en p ope ies; he same
ma e ial can ha e di e en composi ions. A p o ocol consis s
o measu emen s o a specific endpoin in gi en condi ions,
ela ed p o ocols o m an in es iga ion en i y; finally, di e en
subs ances can be g ouped in o an assay en i y when he same
p o ocol applies, gi ing an ex emely flexible s uc u e [208].
Wi h he explosi e g ow h o ma e ial da abases, ML ame-
wo ks, and hei success in ma e ial modeling, i is c i ical o
explo e he link be ween he es ima ed ma e ial p ope ies and
expe imen ally measu ed sa e y o unc ional p ope ies. In
NanoIn o maTIX [36], bo h expe imen al da a om selec ed
use cases and also calcula ed desc ip o s a e s o ed in he
eNanoMappe da abase and an e o is de o ed owa ds p o id-
ing open sou ce lib a ies o acili a e in eg a ion wi h da a anal-
ysis amewo ks and de eloping explo a o y da a analysis
me hods. The alida ion o compu a ional models elies on
high-quali y expe imen al da a; such da a may no always be
comple e, and i is necessa y o iden i y da a gaps and, e en u-
ally, o gene a e addi ional da a based on expe imen al and he-
o e ical chemis y and on biology. The goal o his
compu a ional and heo e ical endea o is he ealiza ion o sa e
nanopa icles and nanoma e ials. To ulfil such expec a ions,
heo e ical desc ip o s mus be ansla ed in o measu able
pa ame e s, and his challenge en ails a deep knowledge o he
mode o ac ion o nanopa icles ha lead o ha m ul ou comes.
On ano he le el, iden ifica ion and es ima ion o heo e ical d i-
e s o oxici y can acili a e he p io i iza ion o expe imen al
es s, which, due o he uncon ollable inhomogenei y o any
ensemble o nanopa icles, is always needed o a obus isk
assessmen o nano-enabled echnology.
G ouping app oaches
Regula o y p ocesses, which o en ely on in i o es ing, a e ou -
paced by he inc easing numbe o ENMs on he ma ke . To cope
wi h his si ua ion, he lack o da a and o ensu e he sa e y o
new ma e ials, g ouping app oaches eme ge as an in e es ing
me hod. Those app oaches a e accep ed wi hin he o e a ching
EU chemicals egula ion REACH (EC 1907/2006) and a e com-
monly used o conside mo e han one chemical a he same
ime [209–211]. Wi hin an es ablished g oup, da a gaps can be
filled by ead-ac oss. He e, exis ing da a on a pa icula (eco) ox-
icological endpoin linked o one o se e al sou ce chemicals can
be employed o es ima e he same p ope y o one o mo e a ge
chemical(s).
Chemicals can be g ouped on he basis o well-defined physic-
ochemical simila i ies, like common unc ional g oups, p ecu -
so s, and/o b eakdown p oduc s. Howe e , ENMs pose an
addi ional challenge compa ed o chemicals, since he e is a e y
limi ed unde s anding o how indi idual physicochemical
pa ame e s influence cellula up ake and oxici y. In addi ion,
he p ope ies o ENMs can change depending on he su ound-
ing medium and o e ime. To es ablish g ouping app oaches o
ENMs, i is essen ial o unde s and how he indi idual physico-
chemical p ope ies a e linked o oxici y. To suppo a g ouping
jus ifica ion, addi ional in o ma ion on a common Mode o
Ac ion (MoA) o oxici y mechanism is ad an ageous [209].
The MoA o a subs ance desc ibes he unc ional o physiological
FIG. 8
eNanoMappe /Ambi da a model adap ed om Re . [208]. Subs ances a e cha ac e ized by hei composi ion and iden i ied by name and ID, which can be
mul iple. Complex ela ions be ween he subs ance componen s can be speci ied.
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
364

changes i causes o a li ing o ganism o cell. One o he d aw-
backs is ha oxici y mechanisms a e only pa ially unde s ood,
and o plen y o ENMs a ian s, he p ecise MoA emains elu-
si e. In his ega d, he po en ial o sys ems biology o con ibu e
o he de elopmen o eliable g ouping app oaches o ENMs
should no be obli e a ed. Mode n omics-based app oaches ( an-
sc ip omics, p o eomics, me abolomics), in combina ion wi h
sophis ica ed bioin o ma ics and da a analysis ools, a e e y
impo an o cha ac e ize oxici y pa hways and un a el ela-
ionships be ween indi idual physicochemical p ope ies and
cellula esponses. This knowledge can hen be applied in he
con ex o g ouping and ca ego iza ion. Mo eo e , omics
app oaches a e o ele ance o he de elopmen o AOPs and
o es ablish eliable, comp ehensi e es ing s a egies building
on he known MoA [212]. AOPs a e a concep ual cons uc ha
in eg a es known in o ma ion om a ious sou ces in a sequen-
ial chain o causally linked key e en s ha co e di e en le els
o biological o ganiza ion (i.e. cellula , o gan le el) s a ing wi h
a molecula ini ia ing e en leading o he final ad e se ou come
[213]. The knowledge gained om g ouping app oaches can
hen be di ec ly used o SSbD o ENMs. Cu en ly, he e a e se -
e al ENM g ouping amewo ks wi h di e en app oaches [214–
216]. Howe e , he e a e only e y ew case s udies o which
he amewo ks ha e been applied o ENMs.
Recen ly, se e al publica ions ha e aken ad an age o di e -
en echniques, including ML app oaches. [217–219] Addi ion-
ally, he inco po a ion o omics da ase s o suppo he
de elopmen o mo e accu a e g ouping s a egies o ENMs ake
in o accoun he mode o ac ion [220–223].
Bioin o ma ics and ML echniques a e hus essen ial when
app oaching cellula e ec s comp ehensi ely, pa icula ly in
combina ion wi h high- h oughpu echniques like omics.
Fig. 9 depic s he ML andom o es app oach used o he g oup-
ing o ENMs, wi h his s a egy, he biological ac i i y o ENMs
can be p edic ed p o ided ha physicochemical p ope ies a e
known. Omics s udies a e a massi e sou ce o da a se s, which
comp ehensi ely desc ibe he cellula al e a ions caused by any
ea men , impo an ly in his case, by ENMs. Thus, omics me h-
ods a e highly use ul o iden i y he MoA o ENMs o be
employed in he g ouping app oaches as addi ional biological
desc ip o s. So a , mos o he e o s o unde s and he MoA
o ENMs ha e been unde aken in he field o ansc ip omics
[224,225]. Howe e , p o eomics can be e en mo e in o ma i e
since i depic s he cellula al e a ions much close o he pheno-
ype han ansc ip omics. The challenge he e is he s anda diza-
ion o he me hods, pa icula ly o da a analysis and
in e p e a ion [226,227].
Me a-analysis o publicly a ailable p o eome da a a ge ed o
specific o gan al e a ions is being ca ied ou o in es iga e he
MoA o ENMs wi hin he o gan, based on p o eome al e a ion e i-
dence. Rele an da ase s om publicly accessible p o eomics da a-
bases such as PRIDE a e iden ified in he fi s s ep. These da a do
no necessa ily in ol e only ENM ea men s, bu also o he al e -
a ions like disease, cance , and chemical ea men s. A ecen ly
de eloped wo kflow by he B R o s anda dized da a analysis
FIG. 9
Schema ic ep esen a ion o he andom o es app oach used o g ouping o ENMs. ENMs a e desc ibed by a se o di e en physicochemical p ope ies,
and an ac i i y label is assigned o each ENM (based on he ou come o biological assays). ENMs a e hen subjec ed o a andom o es model, in which ENMs
a e di ided in o ac i e and passi e ma e ials, based on spli s made on hei physicochemical p ope ies in each ee. Recu si e ea u e elimina ion is used o
selec only he mos impo an physicochemical p ope ies o achie ing maximum accu acy. To add ess and a oid a la ge o e i ing bias, he model is
alida ed in a lea e-one-ou app oach. This app oach can be used o p edic he biological ac i i y o new ENMs, o which he physicochemical p ope ies
a e known.
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
365
was applied o he da a se s. These esul s a e hen in eg a ed in o
p o eomics da a de no o gene a ed om s udies e alua ing he
e ec o ENMs in i o. Co ela ions be ween nanoma e ials
e ec s and o gan-specific al e a ions can hus be de ec ed.[228]
Concluding ema ks
Gaining a clea and deep a ionale behind he nanosa e y cha ac-
e is ics o ENMs is a mul i ace ed issue s ill posing o midable
challenges: nanoma e ials p esen a mo e complex physico-
chemical p ope ies han hei mac oscopic coun e pa s, ha ing
high su ace eac i i y and he abili y o en e li ing cells, po en-
ially causing damage o cells o en i e o ganisms.
In his wo k, mos ly ocusing on a compu a ional pe spec i e,
we made an e o o e iew and discuss s a e-o - he-a physics-
based models o compu ing bo h in insic and ex insic ENMs
p ope ies ha a e c ucial o se ing up eliable da a-based mod-
els o nanosa e y assessmen . In his spi i , one majo aim o his
wo k was he iden ifica ion o he mos c i ical oadblocks
owa ds compu e -aided suppo o nanosa e y assessmen .
Impo an ly, we ha e iden ified and discussed oppo uni ies in
ad ancing he field and in opening esea ch di ec ions, as con e-
nien ly summa ized in Tables 1 and 2. We ha e ex ensi ely illus-
a ed ecen ly-used me hods o compu ing ad anced
desc ip o s and he cu en associa ed challenges mainly leading
o low da a a iance despi e he expec ed highe fideli y. Hence,
possible sugges ions on hyb idiza ion s a egies o mo ing
owa ds models wi h bo h highe da a a iance and fideli y, as
well as he inclusion o Ad e se Ou come Pa hways (AOPs) a e
en isioned.
We s ess ha he epo ed analysis and sugges ed guidelines
o u u e eflec ou cu en bes unde s anding o he field a e
se e al yea s o discussion among expe s in mul i-disciplina y
ye dispa a e ela ed fields. As such, we hope ha his wo k can
se e as a s imulus o u u e mul idisciplina y esea ch, and
could hus help nuclea e u he b eak h oughs in he compu a-
ional nanosa e y assessmen o ENMs.
Da a a ailabili y
This is a e iew a icle and da a a ailabili y does no apply.
Decla a ion o Compe ing In e es
The au ho s decla e ha hey ha e no known compe ing financial
in e es s o pe sonal ela ionships ha could ha e appea ed o influ-
ence he wo k epo ed in his pape .
Acknowledgemen s
This esea ch has ecei ed unding om he Eu opean
Union’s Ho izon 2020 esea ch and inno a ion p og am unde
g an ag eemen No 814426 (NanoIn o maTIX p ojec ).
TABLE 2
Summa y o ecommenda ions and possible u u e esea ch a enues in he con ex o using ma e ial modeling echniques o compu ing ad anced
desc ip o s.
Modelling echnique Recommenda ions o u u e esea ch
Quan um Compu a ions (DFT and
DFTB)
Those a e among he mos ime-consuming me hods o ex ac ing ad anced desc ip o s. The mos ecen
me hodologies based on machine lea ning algo i hms a e p omising o expand he se o sys ems, bu a en ion
should be paid o hei obus ness and he ene ge ic needs associa ed wi h massi e calcula ions. Plu idisciplina
physico-chemical app oaches, such as hose widely used in o he echnological fields like ca alysis, should be used
on a egula basis, o cap u e oxici y mechanisms on he molecula le el. This may lead o new desc ip o s and
inc ease he p edic i e powe .
Reac i e A omis ic Simula ions Sui able eac i e o ce fields a e s ill o be de eloped o accu a ely p edic ing he de elopmen o ENMs su ace
elec ical cha ges. In his espec , eaxFF [191] o he mo e ecen Machine Lea ning based po en ials [192] a e
expec ed o likely play an impo an ole in coping wi h su ficien ly la ge pa icles beyond he capaci y o s anda d
DFT and DFTB compu a ions.
Classical All-A om Molecula
Dynamics (AAMD)
Calcula e he adso p ion a fini y o small me aboli es on nanopa icles using enhanced sampling echniques such
as me adynamics. Molecula nano-desc ip o s in [124] ha e p o ed e y compu a ionally e ec i e o cope wi h
la ge size pa icles, none heless hey a e limi ed o non-me allic pa icles: Addi ional e o should be spen o elax
his cons ain . AAMD is also needed o he molecula docking o p o eins o ENM su aces (see below). To speed
up u u e calcula ions, an MD da abase o ypical con o ma ions and hei co esponding he modynamic weigh s
could be p e-buil o p o eins ha play a majo ole in in e ac ing wi h ENMs (i.e., albumin, mucin, e c.) o inc ease
he accu acy o docking-based calcula ions o nano-desc ip o s (e.g., SASA
H
).
Molecula P o ein Docking Cu en docking so wa e is op imised o p o ein-d ug and p o ein–p o ein in e ac ions (PPIs). Docking
app oaches ely on ini ial AAMD s udies (see abo e) o he p o eins used. Fu u e esea ch would benefi om
op imiza ion o sampling e ficien ly and accu a ely p o ein-ENM su aces (i.e., ino ganic ma e ials). Fu u e docking
p og ams should be able o handle mul iple molecules, including la ge p o eins wi h di e se con o ma ions and
in a much la ge numbe han cu en ly possible (i.e., om ens o hund eds o e en housands). One possible
app oach o b idge he gap be ween he limi a ions o docking p og ams and he complex and la ge-scale na u e
o PPIs and p o ein-ENM in e ac ions is he de elopmen o ad anced da a-d i en and machine lea ning
app oaches.
Coa se-G ained Molecula
Dynamics (CGMD)
Es ima ing abso p ion ene gy densi ies by AAMD o p e-sc eening by con inuum Hel ich me hods. Calcula ing CG
PMFs o all possible co e–shell combina ions o gene a e a ma ix o mapping ma e ial-specific a omis ic PMFs.
When add essing ex insic desc ip o s o ENMs-cell memb ane binding, o e coming he cu en passi e
configu a ions, by adding many lipid ypes ende ing he cellula memb anes as well as memb ane p o eins and
he cy oskele on.
B ownian Dynamics Simula ing nanopa icles in con ac wi h biological molecules o b idge he gap be ween he molecula modelling
scale and he expe imen al scale. Mo e e ec i e app oaches a e needed o compu e and use PMFs in case o non-
sphe ical o aniso opic pa icles.
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
366
Appendix ALis o physics-based desc ip o s ha can be calcula ed h ough ma e ials modelling echniques
Scale Me hod Ad anced desc ip o Uni s Typical alues o a 3
nm TiO
2
nanopa icle
No es and limi a ions
Quan um - Space scale:
10
11
-10
9
m Time
scale: s a ic
DFT & DFTB S anda d en halpy o o ma ion eV -9 o 8 Ve y accu a e bu ideal only o small
nanopa icles up o 4 nm in diame e .
Highly dependen on he basis se
and he pseudopo en ial o choice.
Absolu e ene gy alues a e only
use ul o compa ison wi h o he
s uc u es.
To al ene gy eV -90000 o 10000
Elec onic ene gy eV 88000
Ene gy o he Highes Occupied Molecula O bi al
(HOMO)
eV 3.2
Ene gy o he Lowes Unoccupied Molecula O bi al
(LUMO)
eV 3.1
HOMO–LUMO ene gy gap eV 0–3
Valence band wid h eV 7.3
Conduc ion band wid h eV 2.6
Fe mi le el eV 3.3
Hyd a ion ene gy eV 143
Ve ical ioniza ion po en ial eV 3.4
Ve ical elec on a fini y eV 3
Oxygen acancy o ma ion ene gy eV 3 o 4
Molecula , a omis ic -
Space scale: 10
9
-
10
8
m - Time scale:
s a ic
Quan um mechanics a
DFT le el
Chemical composi ion (x9) - N.A. Fas o calcula e because he e is no
need o expensi e ene gy
minimiza ion. No o me allic ENMs.
Calculable also o la ge
nanopa icles up o 60 nm. Possible
use in QSAR and ML models. De ailed
desc ip ion o hose quan i ies can
be ound in Re s. [122,124]
Po en ial ene gy (x9) eV -75 o 18
Topology desc ip o s (x9) - 2.5 o 6
Size desc ip o s (x3) Å,Å
2
,Å
3
N.A.
La ice ene gy desc ip o s (x5) eV, eV/Å, eV/Å
2
,
eV/Å
3
-110 o 0
Fo ce field desc ip o s (x27) - N.A.
ReaxFF MD Su ace cha ge C/m
2
-0.06 Need a eac i e o ce field de eloped
o a simila sys em
Pa icle-Memb ane binding ee ene gy eV 0
Agg ega ion ee ene gy kJ/mol 54
Sol en Accessible Su ace A ea (SASA) Å
2
61.27 
s 0.68 No e: See Re . [120].
Molecula scale wi h
a omis ic esolu ion.
Space scale: 10
10
-
10
8
m
AAMD and Molecula
Docking
Hyd ophobic SASA ac ion (SASA
H
) % SASA 33.5–54.5
Pola SASA ac ion (SASA
P
) % SASA 40–50 No e: es ima ed o 4 nm TiO
2
NPs
wi h di e se p o ein composi ions o
(con inued on nex page)
RESEARCH: Re iew
Ma e ials Today
d
Volume 67
d
July/Augus 2023 RESEARCH
367
Re e ences
[1] D. As uc, Chemical Re iews 120 (2020) 461–463.
[2] A. Gizza o e al., Ad anced unc ional ma e ials 24 (29) (2014) 4584–4594.
[3] A. Ca dellini, M. Fasano, E. Chia azzo, e al., Phys. Le . Sec . A Gen. A . Solid
S a e Phys. 380 (20) (2016) 1735–1740.
[4] S.S. Mukhopadhyay, Nano echnol. Sci. Appl. (2014) 63–71.
[5] B.S. Sekhon, Nano echnol. Sci. Appl. (2010) 1–15.
[6] E. Chia azzo, P. Asina i, Nanoscale esea ch le e s 6 (1) (2011) 1–13.
[7] E. Chia azzo, P. Asina i, In e na ional jou nal o he mal sciences 49 (12)
(2010) 2272–2281.
[8] A. Haase, F. Klaessig, EU US Roadmap Nanoin o ma ics 2030 (2018) 1–126.
[9] A. Ca dellini e al., Nanoscale 11 (9) (2019) 3925–3932.
[10] Na . Nano echnol. 16 (6) (2021) 607. h ps://www.na u e.com/a icles/s41565-
021-00911-6.
[11] Y. Zhang, C. Ling, npj Compu . Ma e . 4 (1) (2018) 28–33.
[12] Communica ion om he commission o he eu opean pa liamen , he
council, he eu opean economic and social commi ee and he commi ee o
he egions chemicals s a egy o sus ainabili y owa ds a oxic- ee
en i onmen , in: COM/2020/667 Final, 2020.
[13] S.F. Beze a e al., Con ac De ma i is 84 (2) (2021) 67–74.
[14] K. Jagiello, K. Ciu a, Nanoscale 14 (18) (2022) 6735–6742.
[15] K. Jagiello e al., Small 17 (15) (2021).
[16] K. Jagiello e al., En i on. Sci. Nano 9 (5) (2022) 1675–1684.
[17] C. Caldei a, R. Fa cal, C. Mo e i, L. Mancini, H. Rausche , J. Riego Sin es, S.
Sala, K. Rasmussen, Sa e and sus ainable by design chemicals and ma e ials:
e iew o sa e y and sus ainabili y dimensions, aspec s, me hods, indica o s,
and ools, Publica ions O fice o he Eu opean Union, 2022.
[18] B. S iebe o a e al., J. Clean. P od. 241 (2019).
[19] A. Ga cía-Quin e o, M. Palencia, Sci. To al En i on. 793 (2021) 148524.
[20] E. Ma coulaki e al., NanoImpac 23 (2021) 100337.
[21] L.J. Johns on e al., NanoImpac 18 (2020) 100219.
[22] S. Go a do e al., NanoImpac 21 (2021) 100297.
[23] P. Nyma k e al., Small 16 (6) (2020).
[24] S.H. Doak e al., Small 18 (17) (2022).
[25] L. Baja d e al., En i on. Res. 114650 (2022).
[26] S.I. Gomes, J.J. Sco -Fo dsmand, M.J. Amo im, Nano Today 40 (2021) 101242.
[27] C. Wes mo eland e al., Regul. Toxicol. Pha macol. 135 (2022) 105261.
[28] OECD, Guidance Documen on he Valida ion o (Quan i a i e) S uc u e-
Ac i i y Rela ionship [(Q)SAR] Models, 2014.
[29] A. A an i is e al., Compu . S uc . Bio echnol. J. 18 (2020) 583–602.
[30] S. So, J. Rho, Nanopho onics 8 (7) (2019) 1255–1261.
[31] I. Kim e al., Mic ob. Pa hog. 149 (2020) 104290.
[32] NanoSol eIT, NanoSol eIT Ho izon 2020 p ojec , h ps://co dis.eu opa.eu/
p ojec /id/814572.
[33] Go 4Nano, Go 4Nano Ho izon 2020 p ojec , h ps://co dis.eu opa.eu/p ojec /
id/814401.
[34] NANORIGO, NANORIGO Ho izon 2020 p ojec , h ps://co dis.eu opa.eu/
p ojec /id/814530.
[35] RiskGONE, RiskGONE Ho izon 2020 p ojec , h ps://co dis.eu opa.eu/p ojec /
id/814425.
[36] NanoIn o maTIX, NanoIn o maTIX Ho izon 2020 p ojec , h ps://co dis.
eu opa.eu/p ojec /id/814426.
[37] Blekos, K., Ma coulaki, E. (2023). A Bayesian-based sc eening amewo k o
op imal de elopmen o sa e-by-design nanoma e ials. In Compu e Aided
Chemical Enginee ing (in p ess), Else ie .
[38] K. Blekos e al., Jou nal o Chemin o ma ics 15 (1) (2023) 1–17.
[39] H. Nagai, S. Toyokuni, A ch. Biochem. Biophys. 502 (1) (2010) 1–7.
[40] F.S. Bie kand e al., Toxicology esea ch 7 (3) (2018) 321–346.
[41] K. Donaldson e al., Pa icle and fib e oxicology 7 (1) (2010) 5.
[42] A.A. Sh edo a e al., Ame ican Jou nal o Physiology-Lung Cellula and
Molecula Physiology 289 (5) (2005) L698–L708.
[43] M.V. Pa k e al., Bioma e ials 32 (36) (2011) 9810–9817.
[44] D.J. Smi h e al., Appl. Phys. 51 (29) (2018).
[45] R.W. Home e al., J. Chem. In . Model. 48 (12) (2008) 2294–2307.
[46] S.J. Coles e al., O g. Biomol. Chem. 3 (10) (2005) 1832–1834.
[47] I. Lynch e al., Nanoma e ials 10 (12) (2020) 1–44.
[48] E. Wy zykowska e al., Na . Nano echnol. 17 (2022) 924–932.
[49] M. Swi og e al., Sci. To al En i on. 840 (2022) 1–7.
[50] M. F onzi e al., Nanoma e ials 12 (21) (2022).
[51] H. Li e al., J. Phys. Chem. B 127 (15) (2023) 3596–3605.
[52] T. Puzyn e al., Na . Nano echnol. 6 (3) (2011) 175–178.
[53] A.A. To opo e al., Chemosphe e 89 (9) (2012) 1098–1102.
(CONTINUED)
Scale Me hod Ad anced desc ip o Uni s Typical alues o a 3
nm TiO
2
nanopa icle
No es and limi a ions
he NP co ona. Va iance can be
smalle o specific sys ems.
Nega i ely and posi i ely cha ged SASA ac ions
(SASAþ&SASA)
% SASA 6–16
Mesoscopic - Space
scale: 10
9
-10
6
m-
Time scale: 10
9
-
10
6
s
B ownian Dynamics De ia ion om Smouluchowski’s agg ega ion kine ics
heo y
- N.A. Simula ions a e as bu equi e
p e ious AAMD simula ions o ge
he o ce field pa ame e s
Coa se G ained MD Memb ane bending igidi y and o he memb ane
ela ed desc ip o s
- N.A.
RESEARCH: Re iew
RESEARCH Ma e ials Today
d
Volume 67
d
July/Augus 2023
368