1 | Page
PICASSO™: A Human-Fi s F amewo k o Fede ally Manda ed New App oach
Me hodologies
Au ho s: P adip a Ghosh1-4, Sap a shi Sinha1, 3, Cou ney Tindle1, 2, Ma k M. Ga ne 5, Hans Cle e s6-9
A ilia ions:
1Depa men o Cellula and Molecula Medicine, Uni e si y o Cali o nia San Diego, Uni e si y o Cali o nia San Diego, La
Jolla, CA, USA.
2UC San Diego HUMANOID™ Cen e , Uni e si y o Cali o nia San Diego, La Jolla, CA, USA.
3Ins i u e o Ne wo k Medicine, Uni e si y o Cali o nia San Diego, Uni e si y o Cali o nia San Diego, La Jolla, CA, USA.
4Depa men o Medicine, Uni e si y o Cali o nia San Diego, Uni e si y o Cali o nia San Diego, La Jolla, CA, USA.
5Agilen Technologies, Mississauga, ON Canada.
6Hub ech Ins i u e, Royal Ne he lands Academy o A s and Sciences (KNAW) and UMC U ech , he Ne he lands.
7Oncode Ins i u e, he Ne he lands.
8Pha ma, Resea ch and Ea ly De elopmen (pRED) o F. Ho mann-La Roche L d., Basel, Swi ze land.
9The P incess Máxima Cen e o Pedia ic Oncology, U ech , he Ne he lands.
* Co espondence o: p [email protected] (P.G)
ABSTRACT
The b idge o humans is whe e d ug disco e y collapses. New App oach Me hodologies (NAMs), powe ed by
o ganoids, mic ophysiological sys ems, and AI/ML, a e eme ging as he ounda ion o a human- i s d ug disco e y
pa adigm. Ye wi hou s anda ds, mos NAMs emain agmen ed, i ep oducible, and de ached om Phase 3–le el
igo : di e si y, ep oducibili y, and clinically ancho ed endpoin s. PICASSO™ (Pheno ype-In o med Clinical
Abs ac ion o Sys ema ic Simula ion and Ou comes) closes his gap by algo i hmically ancho ing NAMs o la ge,
di e se pa ien coho s. By abs ac ing only he essen ial disease-d i ing ea u es ha align wi h clinical ou comes,
PICASSO™ s ips away i ele an complexi y while en o cing ep oducibili y and clinical ideli y. Ra he han
eplacing animal models, i edeploys hem s a egically when hey e lec human disease mechanisms, sha pening
hei ansla ional alue. Compac in scale ye ancho ed o Phase 3–sized popula ions, NAMs ope a ing wi hin
PICASSO™ become s anda dized, scalable, and capable o ou come-le el p edic ions wi h egula o y-g ade
con idence. PICASSO™ is inclusi e bu disce ning, ans o ming NAMs—whe he algo i hmic, human, o animal
— om bou ique pilo ools in o engines o disco e y, ial design, and p ecision he apeu ics, ushe ing in a uly
human-cen e ed biomedical u u e ha passes he “humanness” es .
2 | Page
GRAPHIC ABSTRACT
F om an e a o ep oducibili y c ises oo ed in animal model–based esea ch ( op; adap ed om Declan Bu le , Na u e
20081) o a u u e ha is end- o-end human (bo om). Th ough delibe a e abs ac ion, PICASSO™ in eg a es compu a ional
and sys ems biology app oaches o ans o m NAMs om agmen ed pilo models in o s anda dized, simple, scalable, and
human- i s amewo ks. Ancho ed o la ge, di e se pa ien coho s, and o ounda ional models o cellula in elligence,
NAMs ope a ing wi hin PICASSO™ a e capable o p edic ing clinical ou comes wi h egula o y-g ade con idence.
3 | Page
In B ie :
• Regula o y momen um: FDA Mode niza ion Ac and NIH ini ia i es a e accele a ing adop ion o NAMs as
c edible al e na i es o animal es ing.
• Ecosys em eali y: D ug disco e y is now a connec ed con inuum; NAMs mus unc ion seamlessly ac oss
academia, bio ech, pha ma, and egula o s.
• Scien i ic p omise s. alida ion gap: O ganoids, MPS, and AI models deli e human ele ance bu demand
igo ous in eg a ion, s anda diza ion and egula o y alignmen , pa icula ly on SOPs, measu emen pa ame e s
and e alua ion c i e ia.
• Academic impe a i e: Rep oducibili y and ansla ional benchma king a e essen ial o c edibili y.
• Bio ech impe a i e: Pla o m builde s mus design wi h egula o y end-use in mind, no jus inno a ion, o b idge
disco e y and app o al.
Highligh s:
• FDA and NIH a e no longe asking o NAMs— hey’ e equi ing hem.
• O ganoids, MPS, and AI a e p oposed as eplacemen s o animal models; no all scien is s ag ee.
• The bigges ba ie (s) a e no science; i ’s he lack o alida ion consensus and egula o y cla i y.
• Academia mus mo e beyond one-o “model pape s” and deli e ep oducibili y a scale.
• Bio ech ha does no design NAM pla o ms wi h egula o y end-use in mind will be le behind.
• Real p og ess hinges on b eaking silos; academia, indus y, and egula o s mus co-de elop he playbook.
• Only NAMs (algo i hmic, human and animal models) ha may ely on smalle n o humanness and di e si y han
Phase 3 ials ye emain ancho ed in Phase 3–sized popula ions h ough abs ac ion—and ha can ecapi ula e
uni e sal, measu able, scalable, and egula o - eady pa e ns—will de ine he u u e o ansla ional science.
4 | Page
T ansla ional modeling in biomedicine has long s uggled o b idge expe imen al p edic ions wi h clinical ou comes, ueling
cos ly ailu es and ep oducibili y c ises. The ede ally manda ed emedy—New App oach Me hodologies (NAMs)—is no
longe op ional. NAMs a e apidly becoming cen al o he u u e o d ug de elopmen and chemical sa e y assessmen .
Recen accele an s— he FDA Mode niza ion Ac (2022)2, and he e y ecen FDA ‘Roadmap o Reducing Animal Tes ing
in P eclinical Sa e y S udies’, NIH’s s a egic ini ia i es, and he EPA’s oadmap o educing animal es ing—ha e aligned
wi h a wa e o public and policy p essu es. Sho ly a e he FDA oadmap, he NIH announced hei own ini ia i e (OPRIVA,
he O ice o Resea ch Inno a ion, Valida ion, and Applica ion) o p omo e
he use o NAMs o eplace animal esea ch. A he same ime,
b eak h oughs in o ganoids, mic ophysiological sys ems (MPS), and AI/ML
make i possible o cap u e human biology a apidly inc easing esolu ion.
The con e gence o need and capabili y se s he s age o a
ans o ma ion. The g ow h o esea ch using NAMs has been
ex ao dina y: while in 1981–1984 ewe han 0.05% o NIH g an s
suppo ed animal al e na i es, by 2024 ha numbe had su ged o nea ly
8%—a >150- old inc ease ha signals NAMs’ ise om inge concep o
mains eam engine o biomedical disco e y.
1. De ini ion o New App oach Me hodologies (NAMs)
NAMs is a collec i e e m used by egula o y agencies and he scien i ic
communi y o desc ibe human- ele an app oaches ha can educe, e ine,
o eplace he use o animals in biomedical esea ch, oxicology, and d ug
de elopmen (Box 1). Unlike adi ional models ha ely on e olu iona y
p oxies, NAMs aim o di ec ly in e oga e human biology wi h ools ha a e
mechanis ically p ecise, scalable, and compu a ionally in eg a i e. They
a e no de ined by a single echnology, bu by a philosophy: end- o-end
human sys ems ha can c edibly in o m egula o y, clinical, and
ansla ional decision-making.
The essen ial cha ac e is ics o a NAM a e (Figu e 1):
1. End- o-end human — he biological ma e ials a e, o he ex en
possible, de i ed om humans (e.g., p ima y cells, iPSCs,
o ganoids, enginee ed issues, o human- ele an biochemical
sys ems).
2. Coho -le el measu emen — he pla o ms enable in-dep h,
ep oducible measu emen o clinically ele an pheno ypes ac oss
s a is ically meaning ul sample sizes.
3. Compu a ional in eg a ion — ad anced da a
p ocessing, analysis, and in eg a ion a e embedded,
using a i icial in elligence (AI) and sys ems modeling
o cap u e eme gen p ope ies and p edic ou comes.
BOX 1: De ining New
App oach Me hodologies
(NAMs)
Co e cha ac e is ics o NAMs:
•End- o-end human: Biological
ma e ials de i ed om humans
(p ima y cells, iPSCs, o ganoids,
enginee ed issues, human- ele an
biochemical sys ems).
•Coho -le el measu emen : Abili y o
deli e ep oducible, clinically ele an
pheno ypes ac oss s a is ically
meaning ul sample sizes.
•Compu a ional in eg a ion: Buil -in
use o AI, ML, and sys ems modeling
o analyze and p edic eme gen
ou comes.
Modali ies included unde NAMs:
•3D cul u e sys ems: O ganoids,
umo oids, o gan-on-chip.
•Cell- ee (in chemico) assays:
Biochemical o molecula in e ac ion
pla o ms.
•In silico models: Algo i hmic and
compu a ional simula ions o disease
and he apeu ic esponse.
Key akeaway: NAMs a e no a echnology
class, bu a human-cen ic philosophy o
building egula o - eady models.
Cu en ly, NAMs a e deployed only in
pilo s, no in pi o al p og ams, because
hey lack he Phase 3–g ade me ics ha
ancho egula o y app o al.
5 | Page
While many NAMs a e associa ed wi h 3D cul u e sys ems such as o ganoids, umo oids, o o gan-on-chip models, he
ca ego y is b oade . I also encompasses cell- ee (in chemico) sys ems ha es biochemical o molecula in e ac ions
di ec ly, as well as algo i hmic (in silico) models ha simula e disease p ocesses and he apeu ic esponses. The uni ying
p inciple is no o m, bu unc ion: NAMs p o ide human-cen ic, mechanis ically ai h ul abs ac ions ha can no only se e
as ounda ional essels o ansla ion bu also scale owa d egula o y-g ade decision-making.
2. The Cu en S a e o NAMs: A B ie His o y
The mo emen owa d human- ele an models began mo e han ou decades ago a Johns Hopkins Uni e si y wi h he
launch o he Cen e o Al e na i es o Animal Tes ing (CAAT) in 1981. I s mission, he 3Rs: educ ion, e inemen , and
eplacemen o animal esea ch, ma ked he i s coo dina ed global e o o e hink p eclinical science. A ha ime,
modeling human biology ou side he body was ba ely concei able.
Th ee decades la e , b eak h oughs in adul s em-cell cul u e ede ined wha was possible. I is no ewo hy ha
CAAT was o med a a ime when human model esea ch was no a legi ima e science. 3D epi helial o ganoid app oaches
we e pionee ed by Mina Bissell3. We imp o ed his Ma igel-d i en o ganoid cul u ing app oach4, i s in mice and hen in
humans, such ha issue s em-cell–de i ed epi helial s uc u es we e gene ically s able, expandable, c yop ese able, and
could ai h ully ecapi ula e na i e issue a chi ec u e. Simul aneously, Yoshiki Sasai de eloped app oaches o c ea e cen al
ne ous sys em o ganoids (e.g., b ain, e ina) by di ec ed de elopmen o plu ipo en s em cells, such as iPSCs5.
The comme cial and academic momen um ha ollowed was explosi e. New media o mula ions, sca olds, and co-
cul u e sys ems ans o med o ganoids in o e sa ile new app oach me hodologies (NAMs) o disease modeling, e icacy
es ing, and pe sonalized medicine.
Today, NAMs-- encompassing o ganoids, mic ophysiological sys ems, in chemico, and in silico models—ha e
e ol ed in o a co e pilla o human- ele an science. O ganoids, o gan-on-chip sys ems, high-con en imaging, and AI/ML
pla o ms ha e ans o med he echnical landscape, while OECD amewo ks, FDA quali ica ion pa hways, and NIH c oss-
agency p og ams ha e begun o se s anda ds. The FDA Mode niza ion Ac 2.0 elimina ed he equi emen o animal
es ing be o e IND submission, signaling openness o al e na i es. The FDA’s 2025 oadmap o minimize animal aims o go
u he , manda ing NAM deploymen in he apeu ic monoclonal an ibody p og ams and incen i izing sponso s who submi
NAM da a in pa allel wi h animal da a by o e ing “ egula o y elie ,” such as smalle equi ed animal coho s o educed
p ima e oxicology s udies. Pilo p og ams such as ISTAND [Inno a i e Science and Technology App oaches o New
D ugs] a e designed o accele a e quali ica ion pa hways, while NIH’s NCATS is d i ing c oss-agency in es men s o expand
alida ion pipelines.
Ye , despi e ema kable p og ess, he ield s ill lacks a uni e sal s anda d o benchma king ideli y, scalabili y, and
mechanis ic p edic i i y; no does i ha e Phase 3–g ade me ics ha ancho egula o y app o al, a gap he PICASSO
amewo k aims o ill. NAMs emain incomple e, inconsis en ly alida ed, a ely ha e de ailed demog aphic in o ma ion
abou he sou ce pe son, and poo ly s anda dized, isking p oducing insigh s ha may be elegan in design bu agile in
ansla ion. The e is a ga he ing consensus ha wi hou popula ion-scale da ase s, ep oducibili y ac oss labs, o accep ed
isk amewo ks, NAMs canno ye se e as he bed ock o egula o y con idence.
Figu e 1. Essen ial cha ac e is ics o ‘ideal’ NAMs. The wo k low mus begin [START] wi h Phase 3–sized human coho s, which a e
mined o dis ill disease-d i ing essen ials, s ipping away supe luous complexi y o e eal he p ocesses and a ge s wo h he apeu ic
6 | Page
e e sal. These essen ials a e hen modeled in (p e e ably) p ospec i e “li ing” human bio eposi o ies (o ganoids, MPS, p ima y
cells, and pa ien -de i ed mic obes). While hese models use small-n coho s, hey mus emain clinically anno a ed o allow ou come
acking and pe sonalized es ing. Deep pheno yping and mul i-omics, ancho ed by objec i e gene-exp ession sco es, gene a e
ac ionable insigh s ha o m he cen e o he wo k low. These insigh s mus be i e a i ely e ined h ough AI/ML aining o sha pen
ele ance o clinical endpoin s; ideally, hey should also enable in- silico p ospec i e ials. In doing so, alida ion s udies in small-n
NAMs would always be ancho ed back o coho -le el conclusions o main ain ele ance o clinical endpoin s and ensu e igo and
ep oducibili y. Ancho ing mus be achie ed h ough con inuous and objec i e e alua ion o “model s. disease ma ch,” (using
pa ame e s ha mus a oid ep oducibili y c isis o annual upg ades ha a e subjec o knowledge bias, e.g., gene on ology) wi h
e inemen a e e y s age—disco e y, alida ion, and p edic ion. Valida ion pla o ms—including in chemico assays, ex i o pa ien -
de i ed o ganoids (PDOs), mic ophysiological sys ems (MPS), and p ecision-cu issue slices (PCTS), as well as in i o animal
s udies— eed da a back in o compu a ional sys ems models o p ojec ansla ional impac . Such a closed-loop, human-cen ic
amewo k should enable he ep oducible ansla ion o ounda ional cellula and molecula insigh s in o eliable, ele an and
ac ionable he apeu ic a ge s and companion bioma ke s, he eby es ablishing NAMs as a ue end- o-end ounda ion o human- i s
d ug disco e y.
7 | Page
3. The Incon enien T u h o T adeo s: Complexi y Wi hou Rele ance, Benchma king Wi hou Objec i i y
Biomedical science has been seduced by he alse idol o complexi y. Fo decades, we ha e equa ed “mo e” wi h “ u h”:
mo e cell ypes, mo e in ica e ma ices, mo e omics laye s. Bu his dogma has no deli e ed ansla ion, i has deli e ed
agili y. Complexi y wi hou abs ac ion is no ideli y; i is noise disguised as igo .
O ganoid sys ems a e o en enginee ed wi h e e y possible ea u e piled on, in he hope o mimicking na i e
physiology. Ye e e y laye added wi hou pu pose is a liabili y, each one a po en ial sou ce o i ep oducibili y and d i .
Missing cues a e no inhe en ly a al; hei absence o en sha pens he ocus on wha is essen ial. By con as , bloa ed
models obscu e signal, isk eplica ing he i ele an , and lull us in o belie ing we ha e cap u ed “ he eal hing.” The esul
is supe icial biomimic y: models ha look igh bu say li le abou disease d i e s.
A simila ap exis s in gene exp ession analysis. Fo decades, ou knowledge o a gene unc ion has been ea ed
as gold s anda d, e en hough only 8.2% o he human genome is unc ionally cons ained, and only ~1.5% (es ima ed o
be in he ange o 19,587–20,2456-8) encodes p o eins ha a e esponsible o all cellula unc ions9. While AI has p edic ed
many s uc u es10-12, unc ion emains la gely uncha ed-- 30% o all p o eins a e poo ly unde s ood and o much o he
es , knowledge is pa ial. Benchma king NAMs wi h on ology-based (e.g., Gene On ology13 [GO], Pan he 14, DAVID15)
o pa hway en ichmen ools (Gene Se En ichmen Analysis16 [GSEA], Gene Se Va ia ion Analysis17 [GSVA] and
[ssGSEA18], PandaOmics19, Reac ome20, KEGG20, Ingenui y [IPA; h p://www.ingenui y.com], En ich 21) isks
i ep oducibili y because unc ion-biased amewo ks a e
incomple e, dynamic, and misleading, shi ing longi udinally as
blind spo s a e illed22.
The way o wa d is disciplined es ain . Model only wha
ma e s—and whe e models a e buil , apply objec i e, unbiased,
and ma hema ically p ecise amewo ks ha s ip away noise and
ancho indings o in a ian , human- ele an d i e s. This is no
minimalism o i s own sake; i is a delibe a e s a egy o ideli y,
ep oducibili y, and clinical ele ance. Fo example, applying
SARS-CoV-2 o e al lung o ganoids23 c ea ed models ha lacked
clinical p emise (lungs o child en we e ema kably spa ed by he
i us) and lacked ele ance, assessed using objec i e gene
signa u es om he abs ac ed essen ials o adul disease24.
While adul s de eloped dis inc cy opa hies which bo e
semblance o in e s i ial lung diseases25, child en de eloped
sys emic immune s o m ha la gely spa ed he lungs26. The
p inciple o “less is mo e” mus guide NAM design.
E e y exogenous addi ion isks a i ac s ha may
supp ess cell-au onomous niche signals, o misguided ou pu s
unless pa ame ized by essen ials abs ac ed om coho -le el
compa isons.
Figu e 2. Mul idimensional adeo s in NAM-based esea ch and
de elopmen . NAMs mus balance ealism wi h educ ion, inno a ion
wi h ep oducibili y.
8 | Page
A. A he cen e , a balanced scale symbolizes he o e a ching heme o “ adeo s.” Fou domains su ound i , each wi h dis inc
iconog aphy and pai ed ensions:
(1) Scien i ic T adeo s:
- Complexi y s. In e p e abili y: Inc easing cellula and ma ix di e si y may enhance physiological esemblance bu in oduces noise ha
can obscu e mechanis ic insigh s.
- Fideli y s. Scalabili y: High- ideli y o ganoids o en equi e complex p o ocols, limi ing h oughpu and ansla ional u ili y.
- Supe icial Biomimic y s. Func ionali y: Visual esemblance o na i e issue does no gua an ee unc ional ele ance, especially in
disease modeling.
(2) Technical T adeo s:
- S anda diza ion s. Pe sonaliza ion: Pa ien -de i ed models o e indi idualized insigh s bu challenge ep oducibili y ac oss pla o ms.
- 2D s. 3D Cul u es: While 3D sys ems be e mimic issue a chi ec u e, hey a e less ac able o imaging and manipula ion.
- Longe i y s. S abili y: Ex ended cul u e du a ions enable ch onic modeling bu isk pheno ypic d i and ins abili y.
(3) E hical & Economic T adeo s:
- Comme cializa ion s. Dono T us : The mone iza ion o o ganoid echnologies aises conce ns abou dono consen and e hical
s ewa dship27.
- Inno a ion s. Accessibili y: Ad anced pla o ms may be cos -p ohibi i e, limi ing equi able access ac oss esea ch en i onmen s.
(4) Regula o y T adeo s:
- Valida ion s. Speed: Regula o y igo ensu es sa e y and eliabili y bu may delay adop ion o o ganoid-based al e na i es o animal
models.
B. Toge he , he adeo domains in A unde sco e he need o iden i ying an op imal zone o inno a ion a scale, whe ein in en ional
design and disciplined simpli ica ion in o ganoid and MPS sys ems will a o models ha a e no me ely complex, bu s a egically
cons uc ed o yield in e p e able, scalable, and clinically ele an insigh s. I he goal i o hese p e-clinical models o yield be e esul s
in clinical ials, adeo s in ‘TRIALS’ (Y axis) mus be op imized wi h adical simpli ica ion in complexi y, con en and cos (X axis), h ough
disciplined educ ion o essen ials ha can be scaled, s anda dized, and us ed.
The adeo s a e clea (Figu e 2A): complexi y enhances physiological esemblance bu unde mines
in e p e abili y, scalabili y, and ep oducibili y, while o e simpli ica ion sac i ices dep h o h oughpu . This is he
incon enien u h: we a e was ing esou ces, pe pe ua ing ansla ional ailu es, and widening he gap be ween model
p edic ions and clinical ou comes, isking an e a o ep oducibili y c isis wo se han he decades o animal models.
We belie e ha he u u e o NAMs depends no on mul iplying complexi y, bu on ou abili y o iden i y i s and
subsequen ly con e ge upon ( h ough consensus) on an ‘op imal zone’ o inno a ion a scale (Figu e 2B). Tha pa h o
op imiza ion may equi e adical simpli ica ion—disciplined educ ion o essen ials ha can be scaled, s anda dized, and
us ed.
4. The Case o Abs ac ion: Simpli y, bu no o e simpli y
Abs ac ion is o en misunde s ood as o e simpli ica ion. In eali y, abs ac ion is he a o delibe a e educ ion: s ipping
away wha is supe luous o i ele an o expose o a i e a wha is essen ial. P ecision Medicine is exac ly his: iden i y
he p ima y disease d i e (Abelson Ty osine Kinase in CML) and a ge i ( he s o y o Glee ec). Ma hema ics, enginee ing,
and e en a embody his p inciple. Pablo Picasso’s ske ches o a bull showed ha s ipping away de ail can c ys allize
essence a he han e ase i .
In biomedical science, abs ac ion becomes powe ul when g ounded in ma hema ics and machine lea ning. Models
buil his way a e ep oducible— he answe oday will be he same omo ow—ye adap i e, upda ing as new da a eme ge.
This ensu es s abili y wi hou igidi y.
Applied o NAMs, abs ac ion deli e s h ee ad an ages: (1) i s ips away i ele an complexi y wi h i s a endan
consump ion o ime and esou ces while p ese ing exac ness, (2) i enables i e a i e e inemen as da ase s g ow, and
(3) i ancho s models—whe he in silico o o ganoid-based— o disease-d i ing ea u es ied o clinical ou comes. Machine
lea ning illus a es his p inciple: om housands o a iables in umo ansc ip omes, i dis ills a hand ul o d i e
pheno ypes wi h g ea e ideli y han bulk umo models o e loaded wi h noise ha ine i ably d i o e ime.
9 | Page
Abs ac ion is no educ ionism; i is disciplined p ecision. I ans o ms NAMs om elabo a e eplicas in o
s eamlined, egula o - eady ools ha ocus on wha uly changes ou comes. Simply pu , NAMs unde abs ac ion could
be likened o “comp essed algo i hms” ha a e also expec ed o ou pe o m bloa ed eplicas.
5. The PICASSO™ F amewo k o Abs ac ion: The wha and he how
We p opose PICASSO™: Pheno ype-In o med Clinical Abs ac ion o Sys ema ic Simula ion and Ou comes (Box 2;
Figu e 3), a disciplined, algo i hmic end- o-end human amewo k o making NAMs clinically c edible. Inspi ed by he a is ic
p inciple ha less can e eal mo e, PICASSO™ applies delibe a e educ ion— h ough ou s eps o in eg a ed
compu a ional and sys ems app oaches (Figu e 4A) -- o ensu e models cap u e only wha uly d i es disease and he apy.
Figu e 3. F om noise o nuance: PICASSO™ ans o ms NAMs in o human- i s , egula o y- eady ools o d ug disco e y. The
PICASSO™ amewo k o abs ac ing complexi y and p io i izing essen ial, human- ele an insigh s in biomedical esea ch. Inspi ed by
Pablo Picasso’s a is ic educ ion, he panel shows he p og ession om ana omically de ailed o minimalis ep esen a ions o a bull,
symbolizing he dis illa ion o biological complexi y in o essen ial, ac ionable signals. PICASSO codi ies abs ac ion o clinical ancho ing
(la ge di e se coho s o pa ien s) o en o ce Phase 3 igo and p edic i e ideli y using a simple 4-s ep app oach (see Fig 4).
(i) Iden i y essen ials — Modeling wha ma e s.
To iden i y in a ian , disease-d i ing ea u es wi hin complex omic da ase s, PICASSO™ begins wi h algo i hmic
abs ac ion, dis illing biological complexi y in o logical essen ials ha emain s able ac oss issues, species, and disease
s a es. This s age applies machine lea ning o la ge-scale human da ase s spanning heal h- o-disease con inua, cap u ing
16 | Page
in e en ions, and eco e y. PICASSO™ manda es ha such da a be analyzed wi h he same s a is ical igo equi ed in
Phase 3 ials—including haza d a ios, Kaplan–Meie su i al cu es, qua ile shi s, and mul i a ia e eg ession models.
F amewo ks such as CANDiT62 and FORWARD63 exempli y how hese s anda ds ope a ionalize p edic i e modeling by
aligning molecula pheno ypes wi h longi udinal ou comes.
Ancho molecula pheno ypes o eal-wo ld ajec o ies: Biobanked NAMs mus no only ep oduce disease ea u es bu
ack hem o e ime. PICASSO™ en o ces s anda dized pipelines whe e molecula pheno ypes a e sys ema ically linked
o clinical ou comes (e.g., emission, elapse, esis ance o o he ele an endpoin s64) and alida ed agains eal-wo ld
coho dis ibu ions. This ancho ing ans o ms NAMs in o quan i a i e mi o s o popula ion di e si y, enabling consis en
c oss-s udy benchma king and p edic i e ex apola ion o human popula ions.
Digi al wins and mul i-scale compu a ional models as egula o y asse s: Mul i-scale compu a ional amewo ks—including
Boolean ounda ional models55, in eg a ed end o end h ough digi al bioma ke s (now made accessible o biologis s ia he
web-based in ui i e in e ace o COMPASS™53), and dynamic ne wo k simula ions— unc ion as in silico digi al wins ha
ex end NAMs beyond he pe i dish.
These digi al wins can:
• P ojec he apeu ic and diagnos ic ou comes i ually, o ecas ing e icacy, oxici y, and esis ance be o e
human exposu e.
• In eg a e c oss-scale da a (molecula → issue → o gan → sys em → popula ion) using s anda dized, on ology-
ee logic models.
• Benchma k i ual p edic ions agains eal-wo ld clinical da a o quan i y p edic i e accu acy, eliabili y, and
ep oducibili y.
S anda diza ion p inciples o simula ion and digi al wins:
• Rele ance: Digi al models mus eplica e clinically alida ed endpoin s (e.g., su i al, emission, o gan unc ion).
• Reliabili y: Simula ions mus demons a e ep oducible ou pu s unde iden ical pa ame e inpu s ac oss si es and
so wa e e sions.
• Robus ness: Vi ual models mus ole a e biological and compu a ional pe u ba ions, pa ame e noise, and ime-
se ies d i .
• T aceabili y: E e y model i e a ion mus be e sion-con olled, audi able, and anno a ed wi h inpu da a
p o enance and quali y me ics.
• Valida ion pipeline: Simula ion ou pu s mus unde go he same mul i-s ep quali ica ion as expe imen al NAMs—
analy ical alida ion → biological alida ion → clinical co ela ion.
Hyb id Human–Digi al In eg a ion:
PICASSO™ en isions NAMs and digi al wins as ecip ocal mi o s:
• NAMs in o m digi al wins by p o iding eal biological cons ain s and human-de i ed causal logic.
• Digi al wins guide NAM design by p edic ing op imal pe u ba ions and expe imen al endpoin s.
This hyb id eedback loop c ea es a sel -co ec ing, scalable sys em whe e p edic ions a e con inuously upda ed
agains eal-wo ld da a—ad ancing a new gold s anda d o closed-loop ansla ional modeling.
17 | Page
Regula o y s anda diza ion s a emen :
Digi al wins, like NAMs, mus be alida ed as i - o -pu pose p edic i e ins umen s. Thei c edibili y is ea ned h ough
documen ed ep oducibili y, de ined pe o mance h esholds, and anspa en , explainable logic. Regula o s inc easingly
ea alida ed digi al wins as Phase-0 e idence engines, capable o suppo ing adap i e ial design, label expansion, and
mechanism-based sa e y quali ica ion.
This ou -s ep discipline ans o ms NAMs om ad hoc p oxies in o egula o - eady abs ac ions—minimal in o m,
maximal in clinical ele ance. I c ea es a con inuous, human-cen ic disco e y engine whe e da a in o ms models,
models in o m measu emen s, and measu emen s eed p edic i e sys ems, ancho ing hem ac oss small ‘n’ and la ge ‘n’
da ase s (Figu e 4B). I also c ea es g oundwo k o NAM- i s human ials.
In eg a e Essen ials — The Regula o y Roadmap o Human Logic NAMs
PICASSO™ in eg a es modeling, measu emen , and simula ion in o a uni ied amewo k ha con e s NAMs om
expe imen al a i ac s in o egula o y- eady digi al and biological e idence engines (Figu e 3). This in eg a ion en o ces
causali y, compa abili y, and compliance ac oss e e y laye , om pe i dish o pa ien popula ion, ensu ing ha p edic i e
models a e no jus inno a i e bu audi able, s anda dized, and clinically ancho ed.
Uni ica ion o Logic and Measu emen : The Boolean amewo k-based ounda ional model p o ides he logic, explainabili y
and he e o e, con idence in wha NAMs can p edic ; COMPASS™ p o ides he quan i ica ion53. Toge he , hey de ine wha
mus be measu ed and how i mus be measu ed. BoNE’s Boolean implica ion ne wo ks ans o m biological ela ionships
in o ule-based logic, while COMPASS ansla es ha logic in o de e minis ic ac i i y sco es wi h ixed h esholds and
ep oducible a iance es ima es. The combina ion ensu es ha NAM ou comes a e bo h mechanis ically anspa en and
s a is ically igo ous, he eby b idging he his o ical di ide be ween mechanism and me ic.
In eg a ion ac oss Scales and Modali ies:
PICASSO™ en o ces mul i-scale alignmen ac oss ou dimensions o ansla ion:
1. Biological: Models cap u e cellula and molecula d i e s wi h e ained epigene ic and mechanical ideli y.
2. Func ional: Readou s quan i y o gan-le el pe o mance—ba ie unc ion, me abolism, elec ophysiology.
3. Clinical: Ou comes map o pa ien - ele an endpoin s— emission, eg ession, eco e y.
4. Compu a ional: Digi al wins in eg a e hese laye s in o ep oducible, explainable simula ions.
All ou pu s mus con o m o s anda dized me ada a schemas and QC panels, enabling in e -labo a o y
compa abili y and egula o y aceabili y.
S anda diza ion P inciples o In eg a ion
PICASSO™ manda es ha in eg a ed NAM-digi al ecosys ems adhe e o he ollowing egula o y c i e ia:
• Rele ance: Each componen mus demons a e linkage o a clinically alida ed endpoin .
• Reliabili y: C oss-pla o m ep oducibili y and in e -si e conco dance a e equi ed o quali ica ion.
• Robus ness: Sys ems mus wi hs and pe u ba ions, empo al d i , and en i onmen al a iabili y.
• T aceabili y: All da a, algo i hms, and expe imen al condi ions mus be e sion-con olled, anno a ed, and
accessible o audi .
18 | Page
• T anspa ency: Models mus expose hei causal s uc u e and pe o mance bounda ies; no black boxes.
C oss-Valida ion be ween Biological and Digi al Sys ems
PICASSO™ o malizes bi-di ec ional alida ion:
• NAMs supply empi ical cons ain s and biological p io s o digi al wins.
• Digi al wins p edic pe u ba ions, dose anges, and esponse dynamics o e ine NAM expe imen s.
This closed-loop ensu es ha compu a ional in e ence is con inuously checked agains empi ical u h, p oducing
a ha monized con inuum o e idence accep able o egula o s and indus y alike.
Regula o y Con e gence and Quali ica ion Pa hway.
In eg a ed NAM sys ems a e e alua ed h ough a h ee- ie quali ica ion pipeline:
1. Analy ical alida ion — con i m ep oducibili y and ins umen ideli y.
2. Biological alida ion — con i m mechanis ic alignmen wi h human pa hways.
3. Clinical co ela ion — con i m p edic i e alignmen wi h eal-wo ld ou comes.
Each ie equi es documen ed SOPs, s a is ical pe o mance me ics, and coho ancho ing o human da a. Once
alida ed, NAMs and hei digi al wins can be accep ed as Phase-0 egula o y assays o mechanism-based
app o als, adap i e ial design, o pos -ma ke ing su eillance.
Hyb id E idence Gene a ion.
In eg a ion also ede ines how e idence is c ea ed: NAMs gene a e mechanis ic e idence; digi al wins gene a e p edic i e
e idence. Toge he , hey ul ill he e iden ia y equi emen s o explainable AI and da a-d i en egula o y science.
PICASSO™ hus embodies he “3R-plus-1” s anda d o Rele ance, Reliabili y, Robus ness, and Rep oducibili y as a
quan i iable, in e ope able, and anspa en benchma k o he nex gene a ion o human-logic medicine.
Th ough in eg a ion, PICASSO™ ans o ms NAMs, o ganoids, and digi al wins in o a single, s anda dized
con inuum o disco e y and alida ion. I deli e s a egula o y oadmap whe e logic eplaces opaci y, ep oducibili y eplaces
assump ion, and mechanis ic u h eplaces co ela ion. In his u u e, NAMs no longe imi a e li e, bu hey compu e i , wi h
clinical ideli y, ma hema ical p ecision, and egula o y con idence.
A ew he apeu ic a eas ha e al eady exempli ied his p ocess, albei a ea ly s ages. A ecen s udy ackled he long-elusi e
cance s em cell (CSC), long implica ed as he d i e s o elapse and he apeu ic esis ance62. Using a machine lea ning
amewo k—CANDiT (Cance Associa ed Nodes o Di e en ia ion Ta ge ing)— he au ho s iden i ied he apeu ically
exploi able ansc ip omic ne wo ks capable o selec i ely inducing CSC di e en ia ion and dea h. Ancho ed in Phase 3–
sized human umo coho s, his e o began wi h a single “seed” gene, igo ously alida ed ac oss >16,000 pa ien s in 33
independen s udies, and expanded in o a ne wo k ep oducibly con i med in ~4,600 addi ional pa ien s. Abs ac ed in o
PDO-based NAMs (n = 26), hese ne wo ks yielded esponse signa u es ha , when simula ed ac oss 10 independen
coho s (~2,110 pa ien s) o endpoin s o clinical ele ance, p edic ed a ~50% educ ion in elapse and ecu ence. The
selec i i y o his app oach—killing CSCs while spa ing no mal s em and di e en ia ed umo cells—was explained by
molecula mechanisms unco e ed h ough cell-based and in chemico app oaches a decade ago65, 66. This wo k shows how
e en small PDO numbe s, when embedded in he PICASSO™ amewo k, can uel ep oducible, clinically ancho ed
disco e ies ha b idge mechanism o ou come.
19 | Page
Simila ly, he elusi e gu ba ie s a e, long esponsible o Phase 3 ailu es in In lamma o y bowel diseases (IBD),
was dis illed om ~1,600 da ase s in o a simple co e de ec : bioene ge ic collapse d i en by H₂S oxici y, he esul o ailed
de oxi ica ion machine y63. This essen ial insigh eme ged om F.O.R.W.A.R.D (F amewo k o Ou come-based Resea ch
and D ug De elopmen ), a ne wo k-based a ge p io i iza ion pla o m ha ies molecula s a es o clinical ou comes. Buil
and alida ed ini ially as a map o con inuum s a es in IBD on di e se da ase s om ~1500 pa ien s47, and subsequen ly
ained on se en p ospec i e andomized ials ac oss ou biologics (n = 332 pa ien s), F.O.R.W.A.R.D de ined emission
a he molecula le el and p edic ed, wi h ne wo k connec i i y, he likelihood ha a ge ing a gi en molecule would induce
emission. Benchma king agains 210 comple ed ials ac oss 52 a ge s (~200,000 pa ien s), i achie ed 100% p edic i e
accu acy despi e he e ogenei y in d ug mechanisms and ial designs. Single-cell RNA-seq and a p ospec i e PDO biobank
(n = 29 subjec s, 40 PDO lines57) con i med he emission signa u e as epi helium-speci ic and p edic i e o poo ou comes.
By ancho ing small-scale NAMs o Phase 3–sized coho s, F.O.R.W.A.R.D enables in silico Phase 0 ials: de- isking
de elopmen , e i ing shel ed d ugs, and guiding ial design o ea ly e mina ion. I exempli ies how abs ac ion s ips away
noise, isola es essen ials, and ans o ms compac models in o ep oducible, ou come-ancho ed disco e y engines.
In summa y, PICASSO™ is inclusi e bu disce ning: I does no disca d models—algo i hmic, human, o animal—
bu a he ele a es hem o a highe e iden ia y s anda d, applied only when hey pass he humanness es . Wi hin i s
amewo k, e e y expe imen al o compu a ional sys em ea ns i s place by demons a ing measu able alignmen wi h human
biology, clinical ou comes, and causal logic. PICASSO codi ies he i e pilla s o he egula o y qui e [Rele ance, Reliabili y,
Robus ness, Rep oducibili y, and T aceabili y] as he ounda ion o egula o y-g ade NAMs and digi al wins. These
pa ame e s collec i ely de ine he quali ica ion oadmap, om analy ical alida ion o biological e i ica ion o clinical
co ela ion—s anda dizing how NAMs, o ganoids, and digi al wins ad ance owa d egula o y accep ance. Because
egula o s app o e in e en ions g ounded in mechanisms, no mys e ies, PICASSO™ en o ces explainable p edic i i y:
models ha expose hei causal logic, no conceal i in black boxes. By embedding Boolean causali y ( ia BoNE) and
quan i a i e ancho ing ( ia COMPASS™), NAMs become e idence engines ha a e anspa en , ep oducible, and u u e-
p oo o in eg a ion wi h explainable AI. Th ough hese essen ials, PICASSO™ con e s NAMs om expe imen al a i ac s
in o logical, quan i a i e, and egula o y- eady su oga es o human biology. In sho , PICASSO™ ensu es ha e e y model
ea ns, no decla es, i s ideli y o human u h.
6. Why S anda diza ion Ma e s: Making Biology as Rep oducible as Ma hema ics and Enginee ing
Despi e decades o p og ess, NAMs emain agmen ed. O ganoids, mic ophysiological sys ems, and AI models di e no
jus in composi ion and eadou s bu in wha hey claim o ep esen . Wi hou ha monized p inciples, NAMs isk becoming
a bi a y p oxies, demons a ions o echnical ingenui y a he han engines o clinical ansla ion.
S anda diza ion enables in e ope abili y: labs, ins i u ions, and companies wo king om he same playbook can
compa e esul s, pool da ase s, and build us . Wi h a common language o essen ials, NAMs e ol e om bou ique
expe imen s o ep oducible, egula o - eady pla o ms. The esul is e iciency, c edibili y, and accele a ion owa d human
ou comes.
Ma hema ical abs ac ion is ‘connec i e issue’: Disease is educed o a se o essen ials, iden i ied ma hema ically,
econs uc ed biologically, and simula ed compu a ionally. The p ecision o enginee ing ensu es ideli y; he discipline o
ma hema ics ensu es igo . Toge he , hey ans o m biology in o a p edic i e science.
PICASSO™ p o ides he amewo k o s anda diza ion NAMs need. By codi ying abs ac ion, i ensu es ha
models a e delibe a e econs uc ions o disease-d i ing essen ials, no open-ended exe cises in complexi y. This app oach
aligns wi h he FDA Mode niza ion Ac and NIH ini ia i es, which encou age he use o NAMs bu demand igo ,
20 | Page
ep oducibili y, and ele ance. I also aligns wi h he changing imes in which
biology and medicine a e inc easingly b anches o ma hema ics and
enginee ing. The u u e o ansla ional science lies no in desc ip i e
ca aloging o complexi y bu in ma hema ical in e ence and enginee ing
p ecision. PICASSO™ ope a ionalizes his ision. I p o ides a amewo k
whe e e e y model is bo h an abs ac ion and an enginee ing cons uc ,
b idging he gap be ween biological noise and clinical signal.
7. Call o Ac ion: P epa ing academia, bio ech and pha ma
The biomedical sciences now s and a a c oss oads. One pa h chases
complexi y o i s own sake, p oducing models ha d own in noise. The o he
emb aces abs ac ion as a disciplined a —dis illing only wha ma e s o
deli e ele ance. PICASSO™ cha s his pa h (Figu e 3C).
• Academia mus inno a e and emb ace egula o y-g ade igo ,
embed compu a ional and egula o y expe ise in o disco e y, and
ain a new gene a ion luen in bo h biology and ansla ional
science.
• Bio ech mus design pla o ms o scalabili y, obus ness, and
ep oducibili y, no bou ique demons a ions. Ea ly engagemen
wi h FDA and pa icipa ion in quali ica ion p og ams (e.g., ISTAND)
will be key.
• Pha ma mus pa ne ac oss sec o s o ensu e NAMs a e
popula ion- ep esen a i e and aligned wi h egula o y endpoin s.
• Regula o s, in u n, mus incen i ize and suppo sponso s
deploying NAMs, helping o o e come he unde s andable ea o cau ion a ound egula o y change.
• E e yone in ol ed in NAM c ea ion mus ecommi o espec ing pa ien s’ and dono s’ au onomy, ensu ing
in o med consen o all esea ch samples.
Wi h PICASSO™, NAMs e ol e om p oxies o s anda dized, clinically ancho ed abs ac ions capable o ans o ming
d ug disco e y in o a human- i s , egula o - eady science.
8. Fu u e Di ec ions
NAMs should be posi ioned as complemen a y ex ensions o human clinical biology, no siloed s and-ins.
In eg a ion, no eplacemen : NAMs ex end wha clinical samples o animal models canno —longi udinal access,
mechanis ic dissec ion, and expe imen al con ol—while emaining ancho ed o human ou comes.
Big-da a NAMs: High- h oughpu o ganoid and MPS pla o ms, pai ed wi h AI, can gene a e popula ion-scale,
mechanis ically ich da ase s. These will educe dependence on noisy, inconsis en human sel - epo ing (die , sleep,
li es yle) while cap u ing he ull ange o gene ic and physiological a ia ion.
BOX 3: Why s anda diza ion
ma e s
Challenges NAMs ace oday:
•Lack o la ge-scale alida ed
da ase s
•Lack o s anda diza ion o cul u e
p o ocols and condi ions
•Poo ep oducibili y ac oss labs
•F agmen ed SOPs and e e ence
s anda ds
•Limi ed egula o y us
Wha s anda diza ion enables:
•C oss-lab ha moniza ion →
ep oducibili y
•Popula ion-scale ele ance →
s a is ical powe
•Regula o y alignmen →
accele a ed adop ion
•Reduced noise → clea e signal
o clinical ansla ion
Key akeaway: Wi hou s anda diza ion,
NAMs emain bou ique science; wi h i ,
hey become he backbone o
ansla ional medicine.
21 | Page
S anda diza ion: Re e ence ma e ials, ha monized SOPs, and in e ope able da a-sha ing conso ia will be essen ial o
c edibili y and egula o y up ake.
Case s udies as p o ing g ounds: Oncology (PDO-guided d ug esponse67), li e oxicology68 (FDA’s adop ion o MPS-
based p edic ion o inju y), and IBD (molecula sub yping and he apeu ic e icacy assessmen s wi h o ganoids47, 57, 63)
p o ide immedia e es beds o build us and egula o y con idence.
The ajec o y is clea : NAMs mus scale om bou ique, a isanal models in o obus , in e ope able in as uc u es ha
egula o s and indus y alike can ely on.
9. Closing Vision
NAMs now s and a a ipping poin . Regula o y momen um, echnological ma u i y, and socie al demand o human- ele an
science ha e aligned. The winne s in his landscape will no be hose who showcase no el y alone, bu hose who b idge
no el y wi h c edibili y, scaling NAMs in o ep oducible, egula o - eady ools.
The challenge is scalabili y wi h simplici y. “Simple can be ha de han complex”, said S e e Jobs “bu i is wo h
i in he end because once you ge he e, you can mo e moun ains”. Ma k Twain once apologized o a co esponden o
sending a long le e saying, “I didn’ ha e enough ime o w i e sho one”. Bu he a o abs ac ion demands i . PICASSO™
e ames disease no as an o e whelming mosaic o signals bu as a hand ul o c i ical e en s NAMs mus cap u e wi h
igo . Success will come no om p oducing mo e analy ically imp essi e da a, bu om acking he ew signals ha uly
shi pa ien ou comes.
In he u u e, biology and medicine will become p edic i e sciences--b anches o ma hema ics and enginee ing.
NAMs, s anda dized h ough PICASSO™™, will e ol e om bou ique expe imen s o mains eam p ac ice, de ining a new
e a o human- i s d ug disco e y.
Minimal in componen s, maximal in ele ance— ha is he essence o abs ac ion. Tha is he u u e PICASSO™ pain s.
PICASSO™ dis ills disease in o he ew decisi e e en s NAMs mus ack—clinically ancho ed
pheno ypes ha uly mo e he needle o pa ien s. In doing so, PICASSO disciplines NAMs o
ocus only on ou come-de ining e en s.
22 | Page
Acknowledgmen s
The au ho s wish o apologize o no ci ing many impo an publica ions on his opic due o he limi o e e ences
allowed. P.G. is suppo ed by NIH g an s R01-AI141630 and R01-AI55696. PG was also suppo ed by he Leona
M. and Ha y B. Helmsley Cha i able T us and he P opel a Cu e Founda ion. S.S. was suppo ed h ough The
Ame ican Associa ion o Immunologis s (AAI) In e sec Fellowship P og am o Compu a ional Scien is s and
Immunologis s. The iews exp essed a e hose o he au ho s and do no ep esen hose o hei employe s o unde s.
Decla a ion o in e es s
The au ho s decla e no con lic s o in e es .
23 | Page
REFERENCES:
1. Bu le D. T ansla ional esea ch: c ossing he alley o dea h. Na u e. 2008;453(7197):840-2. doi:
10.1038/453840a. PubMed PMID: 18548043.
2. Adashi EY, O'Mahony DP, Cohen IG. The FDA Mode niza ion Ac 2.0: D ug Tes ing in Animals is Rende ed
Op ional. Am J Med. 2023;136(9):853-4. Epub 20230418. doi: 10.1016/j.amjmed.2023.03.033. PubMed PMID: 37080328.
3. Lee GY, Kenny PA, Lee EH, Bissell MJ. Th ee-dimensional cul u e models o no mal and malignan b eas
epi helial cells. Na Me hods. 2007;4(4):359-65. doi: 10.1038/nme h1015. PubMed PMID: 17396127; PMCID:
PMC2933182.
4. Sa o T, V ies RG, Snippe HJ, an de We e ing M, Ba ke N, S ange DE, an Es JH, Abo A, Kujala P, Pe e s PJ,
Cle e s H. Single Lg 5 s em cells build c yp - illus s uc u es in i o wi hou a mesenchymal niche. Na u e.
2009;459(7244):262-5. Epub 20090329. doi: 10.1038/na u e07935. PubMed PMID: 19329995.
5. Ma iani J, Vacca ino FM. B eak h ough Momen s: Yoshiki Sasai's Disco e ies in he Thi d Dimension. Cell S em
Cell. 2019;24(6):837-8. doi: 10.1016/j.s em.2019.05.007. PubMed PMID: 31173711; PMCID: PMC7085937.
6. Gaude P, Michel PA, Zahn-Zabal M, B i an A, Cusin I, Domagalski M, Duek PD, Ga eau A, Gleizes A, Hina d V,
Rech de La al V, Lin J, Niki in F, Schae e M, Teixei a D, Lane L, Bai och A. The neX P o knowledgebase on human
p o eins: 2017 upda e. Nucleic Acids Res. 2017;45(D1):D177-D82. Epub 20161129. doi: 10.1093/na /gkw1062. PubMed
PMID: 27899619; PMCID: PMC5210547.
7. Aken BL, Achu han P, Akanni W, Amode MR, Be nsdo F, Bhai J, Billis K, Ca alho-Sil a D, Cummins C,
Clapham P, Gil L, Gi ón CG, Go don L, Hou lie T, Hun SE, Janacek SH, Jue emann T, Keenan S, Lai d MR, ..., Flicek
P. Ensembl 2017. Nucleic Acids Res. 2017;45(D1):D635-D42. Epub 20161128. doi: 10.1093/na /gkw1104. PubMed
PMID: 27899575; PMCID: PMC5210575.
8. The UniP o Conso ium. UniP o : he uni e sal p o ein knowledgebase. Nucleic Acids Res. 2017;45(D1):D158-
D69. Epub 20161129. doi: 10.1093/na /gkw1099. PubMed PMID: 27899622; PMCID: PMC5210571.
9. Rands CM, Meade S, Pon ing CP, Lun e G. 8.2% o he Human genome is cons ained: a ia ion in a es o
u no e ac oss unc ional elemen classes in he human lineage. PLoS Gene . 2014;10(7):e1004525. Epub 20140724.
doi: 10.1371/jou nal.pgen.1004525. PubMed PMID: 25057982; PMCID: PMC4109858.
10. Jumpe J, E ans R, P i zel A, G een T, Figu no M, Ronnebe ge O, Tunyasu unakool K, Ba es R, Žídek A,
Po apenko A, B idgland A, Meye C, Kohl SAA, Balla d AJ, Cowie A, Rome a-Pa edes B, Nikolo S, Jain R, Adle J, ...,
Hassabis D. Highly accu a e p o ein s uc u e p edic ion wi h AlphaFold. Na u e. 2021;596(7873):583-9. Epub 20210715.
doi: 10.1038/s41586-021-03819-2. PubMed PMID: 34265844; PMCID: PMC8371605.
11. Baek M, DiMaio F, Anishchenko I, Daupa as J, O chinniko S, Lee GR, Wang J, Cong Q, Kinch LN, Schae e
RD, Millán C, Pa k H, Adams C, Glassman CR, DeGio anni A, Pe ei a JH, Rod igues AV, an Dijk AA, Eb ech AC, ...,
Bake D. Accu a e p edic ion o p o ein s uc u es and in e ac ions using a h ee- ack neu al ne wo k. Science.
2021;373(6557):871-6. Epub 20210715. doi: 10.1126/science.abj8754. PubMed PMID: 34282049; PMCID: PMC7612213.
12. Binde JL, Be endzen J, S e ens AO, He Y, Wang J, Dokholyan NV, Op ea TI. AlphaFold illumina es hal o he
da k human p o eins. Cu Opin S uc Biol. 2022;74:102372. Epub 20220416. doi: 10.1016/j.sbi.2022.102372. PubMed
PMID: 35439658; PMCID: PMC10669925.
13. Ashbu ne M, Ball CA, Blake JA, Bo s ein D, Bu le H, Che y JM, Da is AP, Dolinski K, Dwigh SS, Eppig JT,
Ha is MA, Hill DP, Issel-Ta e L, Kasa skis A, Lewis S, Ma ese JC, Richa dson JE, Ringwald M, Rubin GM, She lock G.
Gene on ology: ool o he uni ica ion o biology. The Gene On ology Conso ium. Na Gene . 2000;25(1):25-9. doi:
10.1038/75556. PubMed PMID: 10802651; PMCID: PMC3037419.
14. Mi H, Mu uganujan A, Casag ande JT, Thomas PD. La ge-scale gene unc ion analysis wi h he PANTHER
classi ica ion sys em. Na P o oc. 2013;8(8):1551-66. Epub 20130718. doi: 10.1038/np o .2013.092. PubMed PMID:
23868073; PMCID: PMC6519453.
15. She man BT, Hao M, Qiu J, Jiao X, Basele MW, Lane HC, Imamichi T, Chang W. DAVID: a web se e o
unc ional en ichmen analysis and unc ional anno a ion o gene lis s (2021 upda e). Nucleic Acids Res.
2022;50(W1):W216-W21. doi: 10.1093/na /gkac194. PubMed PMID: 35325185; PMCID: PMC9252805.
16. Sub amanian A, Tamayo P, Moo ha VK, Mukhe jee S, Ebe BL, Gille e MA, Paulo ich A, Pome oy SL, Golub
TR, Lande ES, Mesi o JP. Gene se en ichmen analysis: a knowledge-based app oach o in e p e ing genome-wide
exp ession p o iles. P oc Na l Acad Sci U S A. 2005;102(43):15545-50. Epub 20050930. doi: 10.1073/pnas.0506580102.
PubMed PMID: 16199517; PMCID: PMC1239896.
17. Hänzelmann S, Cas elo R, Guinney J. GSVA: gene se a ia ion analysis o mic oa ay and RNA-seq da a. BMC
Bioin o ma ics. 2013;14:7. Epub 20130116. doi: 10.1186/1471-2105-14-7. PubMed PMID: 23323831; PMCID:
PMC3618321.
18. Chen Y, Feng Y, Yan F, Zhao Y, Zhao H, Guo Y. A No el Immune-Rela ed Gene Signa u e o Iden i y he Tumo
Mic oen i onmen and P ognose Disease Among Pa ien s Wi h O al Squamous Cell Ca cinoma Pa ien s Using ssGSEA:
A Bioin o ma ics and Biological Valida ion S udy. F on Immunol. 2022;13:922195. Epub 20220706. doi:
10.3389/ immu.2022.922195. PubMed PMID: 35935989; PMCID: PMC9351622.
24 | Page
19. C o D, O'Kelly G, Wu G, Haw R, Gillespie M, Ma hews L, Caudy M, Ga apa i P, Gopina h G, Jassal B, Jupe S,
Kala skaya I, Mahajan S, May B, Ndegwa N, Schmid E, Shamo sky V, Yung C, Bi ney E, ..., S ein L. Reac ome: a
da abase o eac ions, pa hways and biological p ocesses. Nucleic Acids Res. 2011;39(Da abase issue):D691-7. Epub
20101109. doi: 10.1093/na /gkq1018. PubMed PMID: 21067998; PMCID: PMC3013646.
20. Kanehisa M, Go o S. KEGG: kyo o encyclopedia o genes and genomes. Nucleic Acids Res. 2000;28(1):27-30.
doi: 10.1093/na /28.1.27. PubMed PMID: 10592173; PMCID: PMC102409.
21. Kulesho MV, Jones MR, Rouilla d AD, Fe nandez NF, Duan Q, Wang Z, Kople S, Jenkins SL, Jagodnik KM,
Lachmann A, McDe mo MG, Mon ei o CD, Gunde sen GW, Ma'ayan A. En ich : a comp ehensi e gene se en ichmen
analysis web se e 2016 upda e. Nucleic Acids Res. 2016;44(W1):W90-7. Epub 20160503. doi: 10.1093/na /gkw377.
PubMed PMID: 27141961; PMCID: PMC4987924.
22. Phan A, Joshi P, Kadelka C, F iedbe g I. A longi udinal analysis o unc ion anno a ions o he human p o eome
e eals consis en ly high biases. Da abase (Ox o d). 2025;2025. doi: 10.1093/da abase/baa 036. PubMed PMID:
40338520; PMCID: PMC12060720.
23. Lame s MM, an de Vaa J, Knoops K, Riesebosch S, B eugem TI, Myky yn AZ, Beume J, Schippe D,
Bezs a os i K, Koopman CD, G oen N, Ra elli RBG, Duimel HQ, Demme s JAA, Ve jans GMGM, Koopmans MPG,
Mu a o MJ, Pe e s PJ, Cle e s H, Haagmans BL. An o ganoid-de i ed b onchioal eola model o SARS-CoV-2 in ec ion
o human al eola ype II-like cells. EMBO J. 2021;40(5):e105912. Epub 20210111. doi: 10.15252/embj.2020105912.
PubMed PMID: 33283287; PMCID: PMC7883112.
24. Tindle C, Fulle M, Fonseca A, Tahe i S, Ibeawuchi SR, Beu le N, Ka ka GD, Clai e A, Cas illo V, He nandez M,
Russo H, Du an J, C o y Alexande LE, Tipps A, Lin G, This le hwai e PA, Cha opadhyay R, Roge s TF, Sahoo D, ...,
Das S. Adul s em cell-de i ed comple e lung o ganoid models emula e lung disease in COVID-19. Eli e. 2021;10. Epub
20210813. doi: 10.7554/eLi e.66417. PubMed PMID: 34463615; PMCID: PMC8463074.
25. Sinha S, Cas illo V, Espinoza CR, Tindle C, Fonseca AG, Dan JM, Ka ka GD, Das S, Sahoo D, Ghosh P.
COVID-19 lung disease sha es d i e AT2 cy opa hic ea u es wi h Idiopa hic pulmona y ib osis. EBioMedicine.
2022;82:104185. Epub 20220720. doi: 10.1016/j.ebiom.2022.104185. PubMed PMID: 35870428; PMCID: PMC9297827.
26. Ghosh P, Ka ka GD, Shimizu C, Kim J, Khandelwal S, T emoule AH, Kanegaye JT, G oup PEMKDR. An
A i icial In elligence-guided signa u e e eals he sha ed hos immune esponse in MIS-C and Kawasaki disease. Na
Commun. 2022;13(1):2687. Epub 20220516. doi: 10.1038/s41467-022-30357-w. PubMed PMID: 35577777; PMCID:
PMC9110726.
27. de G oo H, an Daal M, Ho land RW, B ons eld I, Jongsma KR, Ten Ham RMT. The e hics and economics o
o ganoid comme cializa ion: po en ial dono s' pe spec i es. BMC Med E hics. 2025;26(1):109. Epub 20250729. doi:
10.1186/s12910-025-01269-3. PubMed PMID: 40730998; PMCID: PMC12309163.
28. Sinha S, Ghosh P. F om Complex o Simple and Back: Ma hema ical Abs ac ion o Li e's Logic Re eals Disease-
D i ing Essen ials. 2025.
29. Cu ius K, W igh NA, G aham TA. E olu ion o P emalignan Disease. Cold Sp ing Ha b Pe spec Med.
2017;7(12). Epub 20171201. doi: 10.1101/cshpe spec .a026542. PubMed PMID: 28490542; PMCID: PMC5710095.
30. Ge s ung M, Jolly C, Leshchine I, Den o SC, Gonzalez S, Roseb ock D, Mi chell TJ, Rubano a Y, Anu P, Yu K,
Ta abichi M, Deshwa A, Win e singe J, Kleinheinz K, Vázquez-Ga cía I, Haase K, Je man L, Sengup a S, Macin y e G,
..., Conso ium P. The e olu iona y his o y o 2,658 cance s. Na u e. 2020;578(7793):122-8. Epub 20200206. doi:
10.1038/s41586-019-1907-7. PubMed PMID: 32025013; PMCID: PMC7054212.
31. Beason-Held LL, Goh JO, An Y, K au MA, O'B ien RJ, Fe ucci L, Resnick SM. Changes in b ain unc ion occu
yea s be o e he onse o cogni i e impai men . J Neu osci. 2013;33(46):18008-14. doi: 10.1523/JNEUROSCI.1402-
13.2013. PubMed PMID: 24227712; PMCID: PMC3828456.
32. Caselli RJ, Langlais BT, Dueck AC, Chen Y, Su Y, Locke DEC, Wood u BK, Reiman EM. Neu opsychological
decline up o 20 yea s be o e inciden mild cogni i e impai men . Alzheime s Demen . 2020;16(3):512-23. Epub
20200106. doi: 10.1016/j.jalz.2019.09.085. PubMed PMID: 31787561; PMCID: PMC7067658.
33. Ves e gaa d MV, Allin KH, Poulsen GJ, Lee JC, Jess T. Cha ac e izing he p e-clinical phase o in lamma o y
bowel disease. Cell Rep Med. 2023;4(11):101263. Epub 20231107. doi: 10.1016/j.xc m.2023.101263. PubMed PMID:
37939713; PMCID: PMC10694632.
34. E ans-Molina C, Sims EK, DiMeglio LA, Ismail HM, S eck AK, Palme JP, K ische JP, Geye S, Xu P, Sosenko
JM, G oup TDTS. β Cell dys unc ion exis s mo e han 5 yea s be o e ype 1 diabe es diagnosis. JCI Insigh . 2018;3(15).
Epub 20180809. doi: 10.1172/jci.insigh .120877. PubMed PMID: 30089716; PMCID: PMC6129118.
35. Timmons JA, Szkop KJ, Gallaghe IJ. Mul iple sou ces o bias con ound unc ional en ichmen analysis o global -
omics da a. Genome Biol. 2015;16(1):186. Epub 20150907. doi: 10.1186/s13059-015-0761-7. PubMed PMID: 26346307;
PMCID: PMC4561415.
36. Wijesoo iya K, Jadaan SA, Pe e a KL, Kau T, Ziemann M. U gen need o consis en s anda ds in unc ional
en ichmen analysis. PLoS Compu Biol. 2022;18(3):e1009935. Epub 20220309. doi: 10.1371/jou nal.pcbi.1009935.
PubMed PMID: 35263338; PMCID: PMC8936487.
37. Haynes WA, Tomczak A, Kha i P. Gene anno a ion bias impedes biomedical esea ch. Sci Rep. 2018;8(1):1362.
Epub 20180122. doi: 10.1038/s41598-018-19333-x. PubMed PMID: 29358745; PMCID: PMC5778030.
25 | Page
38. Schnoes AM, Ream DC, Tho man AW, Babbi PC, F iedbe g I. Biases in he expe imen al anno a ions o p o ein
unc ion and hei e ec on ou unde s anding o p o ein unc ion space. PLoS Compu Biol. 2013;9(5):e1003063. Epub
20130530. doi: 10.1371/jou nal.pcbi.1003063. PubMed PMID: 23737737; PMCID: PMC3667760.
39. B une JP, Tamayo P, Golub TR, Mesi o JP. Me agenes and molecula pa e n disco e y using ma ix
ac o iza ion. P oc Na l Acad Sci U S A. 2004;101(12):4164-9. Epub 20040311. doi: 10.1073/pnas.0308531101. PubMed
PMID: 15016911; PMCID: PMC384712.
40. De a ajan K. Nonnega i e ma ix ac o iza ion: an analy ical and in e p e i e ool in compu a ional biology. PLoS
Compu Biol. 2008;4(7):e1000029. Epub 20080725. doi: 10.1371/jou nal.pcbi.1000029. PubMed PMID: 18654623;
PMCID: PMC2447881.
41. Be glund AE, Welsh EA, Esch ich SA. Cha ac e is ics and Valida ion Techniques o PCA-Based Gene-
Exp ession Signa u es. In J Genomics. 2017;2017:2354564. Epub 20170206. doi: 10.1155/2017/2354564. PubMed
PMID: 28265563; PMCID: PMC5317117.
42. Alkaabi AM, Abdallah AK. Po olio p ac ices in he p incipal e alua ion p ocess: A quali a i e case s udy. Heliyon.
2024;10(21):e39467. Epub 20241018. doi: 10.1016/j.heliyon.2024.e39467. PubMed PMID: 39524733; PMCID:
PMC11546155.
43. Kong W, Vande bu g CR, Gunshin H, Roge s JT, Huang X. A e iew o independen componen analysis
applica ion o mic oa ay gene exp ession da a. Bio echniques. 2008;45(5):501-20. doi: 10.2144/000112950. PubMed
PMID: 19007336; PMCID: PMC3005719.
44. A gelague R, A nol D, B edikhin D, Delo o Y, Vel en B, Ma ioni JC, S egle O. MOFA+: a s a is ical amewo k o
comp ehensi e in eg a ion o mul i-modal single-cell da a. Genome Biol. 2020;21(1):111. Epub 20200511. doi:
10.1186/s13059-020-02015-1. PubMed PMID: 32393329; PMCID: PMC7212577.
45. A gelague R, Vel en B, A nol D, Die ich S, Zenz T, Ma ioni JC, Bue ne F, Hube W, S egle O. Mul i-Omics
Fac o Analysis-a amewo k o unsupe ised in eg a ion o mul i-omics da a se s. Mol Sys Biol. 2018;14(6):e8124. Epub
20180620. doi: 10.15252/msb.20178124. PubMed PMID: 29925568; PMCID: PMC6010767.
46. Aiba S, González-Blas CB, Moe man T, Huynh-Thu VA, Im icho a H, Hulselmans G, Rambow F, Ma ine JC,
Geu s P, Ae s J, an den Oo d J, A ak ZK, Wou e s J, Ae s S. SCENIC: single-cell egula o y ne wo k in e ence and
clus e ing. Na Me hods. 2017;14(11):1083-6. Epub 20171009. doi: 10.1038/nme h.4463. PubMed PMID: 28991892;
PMCID: PMC5937676.
47. Sahoo D, Swanson L, Sayed IM, Ka ka GD, Ibeawuchi SR, Mi al Y, P anadina a RF, Tindle C, Fulle M, S ec
DL, Chang JT, Sandbo n WJ, Das S, Ghosh P. A i icial in elligence guided disco e y o a ba ie -p o ec i e he apy in
in lamma o y bowel disease. Na Commun. 2021;12(1):4246. Epub 20210712. doi: 10.1038/s41467-021-24470-5.
PubMed PMID: 34253728; PMCID: PMC8275683.
48. Lopez R, Regie J, Cole MB, Jo dan MI, Yose N. Deep gene a i e modeling o single-cell ansc ip omics. Na
Me hods. 2018;15(12):1053-8. Epub 20181130. doi: 10.1038/s41592-018-0229-2. PubMed PMID: 30504886; PMCID:
PMC6289068.
49. Way GP, G eene CS. Ex ac ing a biologically ele an la en space om cance ansc ip omes wi h a ia ional
au oencode s. Pac Symp Biocompu . 2018;23:80-91. PubMed PMID: 29218871; PMCID: PMC5728678.
50. Tian Y, Chen X, Ganguli S, edi o s. Unde s anding sel -supe ised Lea ning Dynamics wi hou Con as i e Pai s.
In e na ional Con e ence on Machine Lea ning; 2021.
51. Ta oni JN, G ayson PC, Hu Q, Eddy S, K e zle M, Me kel PA, G eene CS. Mul iPLIER: A T ans e Lea ning
F amewo k o T ansc ip omics Re eals Sys emic Fea u es o Ra e Disease. Cell Sys . 2019;8(5):380-94.e4. doi:
10.1016/j.cels.2019.04.003. PubMed PMID: 31121115; PMCID: PMC6538307.
52. Lin Y, Dong H, Wang H, Zhang T, edi o s. Bayesian In a ian Risk Minimiza ion. 2022 IEEE/CVF Con e ence on
Compu e Vision and Pa e n Recogni ion (CVPR); 2022 18-24 June 2022.
53. Sinha S, Ghosh P. COMPASS: A Web-Based COMPosi e Ac i i y Sco ing Sys em o Na iga e Heal h and
Disease Th ough De e minis ic Digi al Bioma ke s. doi.o g/10.5281/zenodo.17556273. Zenodo 2025.
54. Sahoo D. The powe o boolean implica ion ne wo ks. F on Physiol. 2012;3:276. Epub 20120723. doi:
10.3389/ phys.2012.00276. PubMed PMID: 22934030; PMCID: PMC3429050.
55. Sinha S, Ghosh P. A Founda ional Model o Biological Logic and Disease. Doi: 10.5281/zenodo.17459815 .
2025.
56. Ch isnandy A, Lu ol MP. An ex acellula ma ix niche sec e ed by epi helial cells d i es in es inal o ganoid
o ma ion. De Cell. 2025. Epub 20250716. doi: 10.1016/j.de cel.2025.06.026. PubMed PMID: 40680738.
57. Tindle C, Fonseca AG, Tahe i S, Ka ka GD, Lee J, Mai y P, Sayed IM, Ibeawuchi SR, Vidales E, P anadina a
RF, Fulle M, S ec DL, Anandacha MS, Pe y K, Le HN, Ea J, Boland BS, Sandbo n WJ, Sahoo D, ..., Ghosh P. A li ing
o ganoid biobank o pa ien s wi h C ohn's disease e eals molecula sub ypes o pe sonalized he apeu ics. Cell Rep
Med. 2024;5(10):101748. Epub 20240926. doi: 10.1016/j.xc m.2024.101748. PubMed PMID: 39332415; PMCID:
PMC11513829.
58. A e sano S, Caiazza C, Caiazzo M. Induced plu ipo en s em cell-de i ed and di ec ly ep og ammed neu ons o
s udy neu odegene a i e diseases: The impac o aging signa u es. F on Aging Neu osci. 2022;14:1069482. Epub
20221220. doi: 10.3389/ nagi.2022.1069482. PubMed PMID: 36620769; PMCID: PMC9810544.
