High-Con en Sc eening images s eaming analysis using he
ST iGen me hodology
Lau a Melga -Ga cía
Da a Science & Big Da a Lab, Pablo de Ola ide Uni e si y
Se ille, Spain
[email p o ec ed]
Da id Gu ié ez-A ilés
Da a Science & Big Da a Lab, Pablo de Ola ide Uni e si y
Se ille, Spain
[email p o ec ed]
C is ina Rubio-Escude o
Depa men o Compu e Science, Uni e si y o Se ille
Se ille, Spain
[email p o ec ed]
Alicia T oncoso
Da a Science & Big Da a Lab, Pablo de Ola ide Uni e si y
Se ille, Spain
[email p o ec ed]
ABSTRACT
One o he echniques ha p o ides sys ema ic insigh s in o biolog-
ical p ocesses is High-Con en Sc eening (HCS). I measu es cells
pheno ypes simul aneously. When analysing hese images, ea u es
like luo escen colou , shape, spa ial dis ibu ion and in e ac ion
be ween componen s can be ound. ST iGen, which wo ks in he
eal- ime en i onmen , leads o he possibili y o s udying ime
e olu ion o hese ea u es in eal- ime. In addi ion, da a s eam-
ing algo i hms a e able o p ocess lows o da a in a as way. In
his a icle, ST iGen (S eaming T iclus e ing Gene ic) algo i hm
is p esen ed and applied o HCS images. Resul s ha e p o ed ha
ST iGen inds quali y iclus e s in HCS images, adap s co ec ly
h oughou ime and is as e han e-compu ing he iclus e ing
algo i hm each ime a new da a s eam image a i es.
CCS CONCEPTS
•In o ma ion sys ems → Clus e ing
;
Da a s eam mining
;
•
Compu ing me hodologies → Gene ic algo i hms
;
• Applied
compu ing → Molecula e olu ion; Imaging;␣
KEYWORDS
Real- ime, T iclus e ing, Gene ic ope a o s, High-Con en Sc een-
ing
1 INTRODUCTION
Nowadays, one o he bigges challenges in biology is unde s anding
genes and hei biological ci cui s. Due o ha , echniques ha
in es iga e comple e cellula p ocesses a e becoming mo e ele an ,
as High-Con en Sc eening (HCS). HCS combines an au oma ed
imaging and analysis o in ac cells exposed o some pe u ba ions
(chemical o genomic) ha al e hei pheno ype [11].
HCS is made by many s eps ha can ake oo much ime i hey
a e no e ec i ely done. On he o he hand, hese days, s eam com-
pu ing ends a e ising, i.e., algo i hms ha eac in he as es way
o p o ide in o ma ion om da a in eal ime. Consequen ly, he
p ocessing and analysis o HCS images in a s eaming en i onmen
could gi e quick in o ma ion om hem.
In [
8
] he T iGen algo i hm is p esen ed as a T iclus e ing algo-
i hm ha disco e s g oups o 3D da ase s h oughou ins ances,
a ibu es and ime. In [
12
] ST iGen algo i hm is in oduced. ST i-
Gen is a new inc emen al lea ning me hod ha inds g oups o
simila beha iou pa e ns in 3D s eam da a con inuously. In his
pape , ST iGen algo i hm is applied o HCS images o ge in o ma-
ion om images in eal- ime.
The a icle is s uc u ed as ollows: ST iGen and HCS me hod-
ologies a e p esen ed in Sec ion 2; he expe imen al se up and he
yielded esul s in Sec ion 3; and inally he conclusions a e in Sec-
ion 4.
2 METHODOLOGY
2.1 ST iGen me hodology
ST iGen is a new inc emen al lea ning me hod ha c ea es iclus-
e s (based on T iGen algo i hm [
8
]) and keeps hem upda ed aking
in o accoun he knowledge om p e ious s eams and upg ading
i s lea ning me hod. ST iGen mee s all 4 Da a S eaming equi e-
men s, i.e., da a has o be p ocessed in he o de o i s a i al and
one by one; he lea ning model
has o be upda ed inc emen ally;
he model has o deal wi h small amoun o memo y e e ing o
he huge quan i y o da a and i has o be as .
ST iGen algo i hm is inspi ed in he o line/online app oach o
s eam algo i hms. Consequen ly, ST iGen s a s wi h an execu ion
o he T iGen modi ied o ea s eam da a as s a ic da a wi h W
da a, whe e W is he maximum numbe o da a s eams ha can
be used in an i e a ion o he ST iGen algo i hm. A e wa ds, he
algo i hm p ocesses each new da a s eam and he model upda es
inc emen ally and quickly basing i sel on he new s eams in o de
o p o ide he upg aded iclus e s.
Du ing his second phase o ST iGen, i ies o ex end he ac ual
iclus e s h oughou ime emo ing he "oldes " ime poin o
keep always a maximum o Wda a poin s, a egula p ocedu e
in Da a S eaming algo i hms [
6
]. In addi ion o he ex ension o
he ac ual iclus e s o e ime, he lea ning model adjus s i sel
inc emen ally depending on he GRQ (GRaphical Quali y) measu e,
pa o one o he T iGen i ness unc ions [7].
Mo e speci ically, he ST iGen lea ning model makes mu a ions
in o iclus e s, i.e., adding, dele ing and/o changing bo h ins ances
and/o a ibu es, o be upda ed. These ope a ions a e quicke han
e- aining T iGen each ime a new s eam a i es o keep he
lea ning model upda ed. Mu a ions allow o ind cu en and global
eal solu ions as ST iGen depends on he esul s om he i s
execu ion o T iGen ha can change e e y ime i is execu ed. In
his way, iclus e s a e mu a ed un il hei GRQ alues a e highe
han minGRQ o un il he numbe o i e a ions made is highe han
numI . In his way, ST iGen ies o include o emo e some cu en
iclus e ’s componen s in o de o keep mos accu a e componen s.
These 2 pa ame e s and also he "window" Wpa ame e and a
minimum GRQ h eshold ( ha can dele e he oldes ime included
in he iclus e when i s GRQ is smalle han his) ake di e en
alues depending on he da ase , o be able o adjus o small o /and
ab up changes in s eams.
When he da ase is syn he ic o when he esul ing iclus e s
a e known in ad ance, a alida ion p ocess o compa e ounded
and eal iclus e s is done wi h accu acy and F1 Sco e measu es.
Da ase s ha a e nei he syn he ic no wi h known iclus e s in
ad ance, a e e alua ed depending on he GRQ alue.
2.2 High-Con en Sc eening me hodology
High-Con en Sc eening o Analysis (HCS o HCA) is gaining im-
po ance. HCS combines au oma ed imaging acquisi ion and image
analysis using, mos equen ly, au oma ed luo escence mic oscopy
[
4
]. Du ing he HCS p ocess in ac cells a e incuba ed wi h sub-
s ances ha al e hei pheno ype in a desi ed way [
3
]. These cells
a e sc eened and mul iple luo escence eadou s a e measu ed in
pa allel. This p ocess p o ides big olume o da a wi h high bio-
logical in o ma ion con en . Subcellula loca ions and luo escence
colou in ensi y du ing di e en complex cellula e en s, in e ms
o space and ime, a e measu ed wi h he au oma ed image analysis
phase o HCS [
5
]. HCS images ha e been use ul o de ec and s udy
DNA, cy okinesis, cell di ision, cell mig a ion, apop osis, mi osis,
and mo e cellula e en s o a ge componen s.
HCS p ocesses in ol e di e en asks as cell p epa a ion and
labelling, image acquisi ion, image analysis and da a managemen
[
9
]. In e ms o da a challenges, HCS has 2 p incipal issues: da a
s o age and da a p ocessing. One o he main in e es ing poin s o
HCS is he simul aneous analysis o images ha equi es a high
compu ing powe and quickly compu e ne wo k connec ions [1].
2.3 ST iGen applica ion o High-Con en
Sc eening images
Applying ST iGen o HCS images allows o disco e he bes ea-
u es o g oup o ge in o ma ion om cells. Ac ual nume ical ea-
u es ex ac ed om HCS images a e: 1) luo escen ma ke colou ;
2) cell componen ’s shape; 3) spa ial si ua ion; 4) dis ibu ion o
Figu e 1: Example o p ep ocessing phase o HCS images
pixel-colou in a cell egion o in e es o s udy in e ac ions o
co-occu ences [
11
]. Mo eo e , e olu ion h oughou ime is added
o hese 4 ea u es when applying ST iGen o HCS.
The ype o images used in his expe imen a e indi idual in ac
cells luo escen mic oscopy-based HCS images. Images a e p o-
cessed in o de o c ea e a da ase ha i s ST iGen equi emen s.
Fi s ly, each RGB image is ans o med in o a 3D ma ix ep esen -
ing he coo dina es o each image pixel in decimal numbe s. In o he
wo ds, alues ha ep esen images colou s a e he pixel-colou
o luo escen -ma ke colou ea u es men ioned abo e. A il e is
passed h ough e e y image o no include any image ha p esen
noise due o op ic abe a ions, mic oscopy issues o e en bad ac-
quisi ion o he image. Secondly, da a is p epa ed o i ST iGen
da ase speci ica ions, i.e., aking in o accoun : de ec ing he colou s
speci ied o analyse and de ec ing a eas limi s ( o a eas wi h mo e
han an use ixed alue). In addi ion, andom alues be ween 0
and 1000 a e added in o de o make ST iGen able o igno e he
backg ound o images. A g aphical example o his me hodology is
in Fig. 1.
3 RESULTS AND DISCUSSION
In his sec ion, he esul s ob ained by he applica ion o he ST iGen
algo i hm o a HCS images da ase a e p esen ed.
ST iGen has been applied o a se o HCS images om [
2
] o
HeLa cells (ce ical cance cells aken om a woman in he 50s ha
can di ide hemsel es an unlimi ed numbe o imes in well p e-
se ed labo a o y condi ions [
10
]). Fo his expe imen , he da ase
selec ed is he one ha p esen s he eac ion o HeLa cells o T ans-
e in ecep o s (usually applied as cance cell a ge because hey
enhances si e-speci ic he apies). The quali y o he esul ing i-
clus e s is e alua ed wi h he GRQ measu e. In addi ion, o his
expe imen , a da ase wi h he desi ed ST iGen ounded iclus-
e s has been c ea ed in o de o compu e F1 Sco e and Accu acy
measu es o check he pe o mance o he algo i hm.
460 HCS images simila o he abo e image in Fig. 1 ha e been
used o his expe imen wi h black o he backg ound image,
blue o he cell bo de , ed o he nuclea DNA and g een o he
endosome compa men . Cell bo de s (blue colou ) has been igno e
because hey do no p o ide ex a in o ma ion abou HCS ea u es.
Table 1: ST iGen con igu a ion pa ame e s
Execu ion minGRQ h esholdGRQ numI
1 0.95 0.80 10
2 0.88 0.70 15
3 0.90 0.75 20
Figu e 2: F1 Sco e esul s
Figu e 3: Accu acy esul s
Figu e 4: ST iGen execu ion ime
ST iGen has been execu ed 3 imes due o he ac ha he i s
iclus e s esul s depend on he i s iclus e s ounded by T iGen.
In ha way, con igu a ion pa ame e s ake di e en alues in each
execu ion o see he in luence o hem in he esul s (see Table 1)
excep ing W ha has as maximum alue 3.
ST iGen pe o mance esul s a e in he ollowing igu es: Fig. 3
shows accu acy alues o all iclus e s and Fig. 2 shows F1 sco e
alues. Figu es show ha he algo i hm pe o ms in an accu a e way
and can ind componen s co ec ly. F1 sco e o he g een iclus e s
in all 3 execu ions a ies a lo , i is due o he ac ha g een a eas
sp ead h ough cells and a e in con inuous mo emen , howe e
he mean accu acy alue in all 3 execu ions is 0.908. Red a eas a e
mo e s able in bo h measu es because hey a e in a simila posi ion
in all s eams.
Apa om hese measu es, ano he impo an pa ame e ha
ep esen s he good pe o mance o he algo i hm is ime execu ion.
This expe imen has been done in a compu e wi h an i7-5820K
3.3GHz p ocesso and 48GB RAM memo y. The i s phase o ST i-
Gen ( ha is jus one execu ion o T iGen wi h he i s 10 s eams)
akes a mean o 8.9 minu es o execu e comple ely. A e wa ds,
he Da a S eaming phase s a s and each new image s eam is
p ocessed and p o ides ounded iclus e s in a mean o 16 seconds.
In gene al, he quali y measu es a e mos ly equal in he 3 ex-
ecu ions. Howe e , i can be seen ha he execu ion ime o he
3 d execu ion o ST iGen is much smalle , due mos ly o he small
alue o numI .
4 CONCLUSIONS
The ST iGen algo i hm pe o ms wi h good esul s when dealing
wi h s eam da a as we ha e seen in Sec ion 3. I is as e han
execu ing he T iGen algo i hm when a new s eam a i es and so
he e olu ion o da a h oughou ime can be analysed in eal- ime.
I p o es ha he lea ning model upda es inc emen ally making
mu a ions and aking in o accoun he Wmos ecen s eams.
The applica ion o HCS images in o he Da a S eaming en i on-
men wi h ST iGen leads he possibili y o ob aining in o ma ion
abou he e olu ion o HCS ea u es, i.e. componen s colou , shape,
loca ion and dis ibu ion, h oughou ime in eal- ime. A possi-
ble applica ion o his expe imen would be he ac ha ST iGen
could de ec when ex e nal agen s like subs ances, d ugs, an ibod-
ies, e c a e added o he exposed cell and how he cell eac s o hem,
e.g., changing iclus e s componen s. I allows o do a con inuous
analysis o he cell in eal- ime.
ACKNOWLEDGMENTS
Au ho s hank he Spanish Minis y o Economy and Compe i i e-
ness o he suppo unde he p ojec TIN2017-88209-C2-1-R and
TIN2017-88209-C2-2-R.
REFERENCES
[1]
M. Bickle. 2008. High-con en sc eening : A new p ima y sc eening ool ? 11, 11
(2008).
[2]
CellO gnize P ojec [n. d.]. CellO ganize p ojec om Ca negie Mellon Uni e -
si y. h p://www.cello ganize .o g/2d-hela/
[3]
D. C onk. 2013. Chap e 8 - High- h oughpu sc eening. (2013), 95 – 117. h ps:
//doi.o g/10.1016/B978-0-7020-4299-7.00008-1
[4]
R. Flaumenha . 2007. 3.07 - Chemical Biology. (2007), 129 – 149. h ps://doi.
o g/10.1016/B0-08-045044-X/00080-8
[5]
G. Galea and J. C. Simpson. 2013. Chap e 17 - High-Con en Sc eening and
Analysis o he Golgi Complex. 118 (2013), 281 – 295. h ps://doi.o g/10.1016/
B978-0-12-417164-0.00017-3
[6]
M. Ghesmoune, M. Lebbah, and H. Azzag. 2016. S a e-o - he-a on clus e ing
da a s eams. Big Da a Analy ics 1, 1 (2016), 1–27. h ps://doi.o g/10.1186/
s41044-016-0011-3
[7]
D. Gu ié ez-A ilés, R. Gi áldez, F.J. Gil-Cumb e as, and C. Rubio-Escude o. 2018.
TRIQ: A new me hod o e alua e iclus e s. BioDa a Mining 11, 1 (2018), 1–29.
h ps://doi.o g/10.1186/s13040-018-0177-5
[8]
D. Gu ié ez-A ilés, C. Rubio-Escude o, F. Ma ínez-Ál a ez, and J.C. Riquelme.
2014. T iGen: A gene ic algo i hm o mine iclus e s in empo al gene exp ession
da a. Neu ocompu ing 132 (2014), 42–53. h ps://doi.o g/10.1016/j.neucom.2013.
03.061
[9]
S. Lee and B. J. Howell. 2006. [25] - High-Con en Sc eening: Eme ging Ha dwa e
and So wa e Technologies. 414 (2006), 468 – 483. h ps://doi.o g/10.1016/
S0076-6879(06)14025-2
[10]
B.P. Lucey, W.A. Nelson-Rees, and G.M. Hu chins. 2009. Hen ie a Lacks, HeLa
cells, and cell cul u e con amina ion. A chi es o Pa hology and Labo a o y
Medicine 133, 9 (2009), 1463–1467.
[11]
F. Heigwe M. Bou os and C. Lau e . 2015. Mic oscopy-based High-con en
Sc eening. Cell 163, 6 (2015), 1314–1325. h ps://doi.o g/10.1016/j.cell.2015.11.007
[12]
L. Melga -Ga cía, D. Gu ié ez-A ilés, and C. Rubio-Escude o. 2019. Disco -
e ing Beha io Pa e ns in Big Da a S eaming En i onmen s : The ST iGen
Me hodology. (2019). Manusc ip submi ed o publica ion.