BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T143140Z
LOCATION:B311
DTSTART;TZID=America/New_York:20241118T151500
DTEND;TZID=America/New_York:20241118T153000
UID:submissions.supercomputing.org_SC24_sess818_ws_cafcw116@linklings.com
SUMMARY:Anti-Cancer Drug Response Prediction on Patient-Derived Xenografts
 : Learning from Extremely Limited Data
DESCRIPTION:Oleksandr Narykov, Yitan Zhu, and Thomas Brettin (Argonne Nati
 onal Laboratory (ANL)); Yvonne Evrard (Frederick National Laboratory for C
 ancer Research); Priyanka Vasanthakumari, Alexander Partin, and Maulik Shu
 kla (Argonne National Laboratory (ANL)); James Doroshow (National Institut
 es of Health (NIH), National Cancer Institute (NCI)); and Rick Stevens (Ar
 gonne National Laboratory (ANL), University of Chicago)\n\nCancer is a fam
 ily of complex genetic disorders characterized by the accumulation of gene
 tic and epigenetic alterations that drive uncontrolled cell growth and met
 astasis. It is becoming a leading cause of death worldwide, accounting for
  10 million deaths in 2020. In 2024, NIH projects that roughly 2 million p
 eople will be diagnosed with cancer in the United States. However, develop
 ing novel therapies or selecting a best-suited treatment for a patient pos
 es a challenge to the scientific community due to the individualized natur
 e of the disease.\n\nCancer research is often hampered by limited data, pa
 rticularly in the context of rare cancer types and heterogenous disease me
 chanisms, restricting the utility of the latest advances in Artificial Int
 elligence (AI) in this domain. This issue is dire for more refined biologi
 cal models, such as patient-derived xenografts (PDX), that are more accura
 te in reflecting patient treatment responses than immortalized cell lines 
 but more costly in terms of time and resources for data generation. Due to
  the high complexity and cost of PDX experiments, it is unlikely to garner
  datasets on the scale of traditional AI domains like computer vision or n
 atural language processing. This situation calls for more efficient approa
 ches to learning from data.\n\nHuman reasoning heavily relies on the conte
 xt surrounding the problem. To achieve generalizability, we compare variou
 s settings and examples to find key differences in subjects and outcomes. 
 This strategy is critical for efficiently utilizing limited available data
  for learning. Currently, the closest analog to this strategy in the AI fi
 eld is Contrastive Learning (CL). This approach leverages the relatively a
 bundant cell line drug screening data for transfer learning to drug respon
 se prediction of PDXs. CL utilizes not only positive data (responsive samp
 les) but also negative data (non-responsive samples), which is ubiquitousl
 y generated during drug response experiments. The most famous example of C
 L is Contrastive Language-Image Pretraining (CLIP), which bridges the gap 
 between textual and visual information. We adopt a similar approach to cre
 ate compatible representations between different biological models, result
 ing in Contrastive Transfer Learning. This emerging machine learning parad
 igm improves the performance and stability of AI models across different a
 pplication domains by emphasizing the learning of more generalizable featu
 re representations. To enhance model explainability, we create a biologica
 lly guided neural network based on the KEGG pathways and BRITE hierarchy s
 o we can elucidate the decision process via its activation pass. \n\nWe ex
 plored the utility of the proposed architecture in drug-specific response 
 modeling, where we constructed an individual response model for each drug 
 based on cancer gene expressions. We used the area under the receiver-oper
 ator curve (AUROC) to evaluate prediction performance. We compare our appr
 oach to the stand-alone fully connected neural network (FCNN) via 5-fold c
 ross-validation (CV). Our approach improves the average AUROC score and re
 duces its standard deviation (std) from CV trials, producing more accurate
  and stable prediction results. Baseline FCNN performance for Selumetinib,
  Bortezomib and Paclitaxel is AUROC=0.87, std=0.188; AUROC=0.92, std=0.068
 ; AUROC=0.87, std=0.077, respectively; while our approach achieves AUROC=0
 .99, std=0.003; AUROC=0.99, std=0.008; AUROC= 0.97, std=0.018 for these th
 ree drugs, correspondingly.\n\nTag: Artificial Intelligence/Machine Learni
 ng, Biology, Broader Engagement, Education, Emerging Technologies, Medicin
 e, Modeling and Simulation\n\nRegistration Category: Workshop Reg Pass\n\n
 Session Chairs: Lynn Borkon (Frederick National Laboratory for Cancer Rese
 arch); Lauren Lewis (Frederick National Laboratory for Cancer Research); a
 nd Eric Stahlberg (MD Anderson Cancer Center, University of Texas)\n\n
END:VEVENT
END:VCALENDAR