BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000711Z
LOCATION:506
DTSTART;TZID=America/Denver:20231112T171500
DTEND;TZID=America/Denver:20231112T173000
UID:submissions.supercomputing.org_SC23_sess430_ws_cafcw112@linklings.com
SUMMARY:Entropy-Based Regularization on Deep Learning Models for Anti-Canc
 er Drug Response Prediction
DESCRIPTION:Oleksandr Narykov, Yitan Zhu, and Thomas Brettin (Argonne Nati
 onal Laboratory (ANL), Data Science and Learning Division); Yvonne Evrard 
 (Frederick National Laboratory for Cancer Research); Alexander Partin, Mau
 lik Shukla, and Priyanka Vasanthakumari (Argonne National Laboratory (ANL)
 , Data Science and Learning Division); James Doroshow (National Cancer Ins
 titute (NCI)); and Rick Stevens (Argonne National Laboratory (ANL))\n\nThi
 s work studies a particular setting for regression problems – tasks with c
 omplex combinatorial data space where samples can be divided into distinct
  groups. Anti-cancer drug response prediction is a perfect example of this
  setting, in which each sample includes cancer biological features and dru
 g chemical information. Many existing works of pan-drug and pan-cancer res
 ponse modeling treat different combinations of drugs and cancers as indivi
 dual samples. A potential problem in these works is that a model may be he
 avily influenced and biased toward overrepresented drugs and cancers. Our 
 work develops a method to solve this issue by adjusting the model training
  process in a deep learning framework. \n\nIn the drug response prediction
  field, the performance of pan-cancer pan-drug models is commonly evaluate
 d on a holdout test set through cross-validation (CV) using performance me
 trics like the coefficient of determination (R2) and the mean squared erro
 r (MSE). However, drug response prediction can be viewed as a multi-object
 ive optimization task, attempting to maximize the prediction performance o
 ver different drugs and cancers. We consider the performance for each drug
  as a separate objective and attempt to find a model on the corresponding 
 Pareto front that provides balanced performances for all compounds. We pro
 pose adding an entropy-based regularization to the loss function for model
  training to reach this balanced state. The intuition behind it is straigh
 tforward – we minimize the MSE on all data points while encouraging the dr
 ug-specific model fitting error variability to stay as low as possible. We
  achieve this by grouping samples by their drug identity and computing the
  MSE for each group in the training batch. Then we calculate entropy over 
 normalized group-specific losses. This value is plugged into the regulariz
 ation term that incentivizes the loss function to maximize it. The maximum
  entropy can be achieved when the MSEs across all drugs take the same valu
 e. \n\nWe investigate the regularization effect on response modeling using
  a drug screening dataset of Cancer Cell Line Encyclopedia (CCLE) and a st
 ate-of-the-art drug response prediction model DeepTTA [1]. We consider two
  CV strategies for model evaluation – random split and drug-blind split. I
 n a random split, the testing set can share both cell lines and drugs with
  the training set, while in a drug-blind split, a drug can not appear simu
 ltaneously in both the training and testing sets. We perform 10-fold CV an
 alyses and evaluate the model performance using R2. For the random split, 
 we see that adding the entropy-based regularization leads to a marginal im
 provement in prediction performance, which is 0.736 without regularization
  versus 0.746 with regularization and a p-value of 0.130 from the pairwise
  t-test. Importantly, we observe a substantial improvement in the more cha
 llenging setting of drug-blind split, where the prediction performance inc
 reases from -0.128 (without regularization) to 0.168 (with regularization)
  with a statistically significant p-value of 0.005 from pair-wise t-test.\
 n\n\n1	Jiang, L., Jiang, C., Yu, X., Fu, R., Jin, S., and Liu, X.: ‘DeepTT
 A: a transformer-based model for predicting cancer drug response’, Briefin
 gs in Bioinformatics, 2022, 23, (3), pp. bbac100\n\nTag: Applications, Sta
 te of the Practice\n\nRegistration Category: Workshop Reg Pass\n\nSession 
 Chairs: Lynn Borkon (Frederick National Laboratory for Cancer Research); S
 ally Ellingson (University of Kentucky); Sean Hanlon (National Institutes 
 of Health (NIH), National Cancer Institute (NCI)); Patricia Kovatch (Icahn
  School of Medicine at Mount Sinai); and Eric Stahlberg (MD Anderson Cance
 r Center, University of Texas)\n\n
END:VEVENT
END:VCALENDAR