Looking at the BiG picture: Incorporating bipartite graphs in drug response prediction

Abstract

Motivation: The increasing number of publicly available databases containing drugs' chemical structures, their response in cell lines, and molecular profiles of the cell lines has garnered attention to the problem of drug response prediction. However, many existing methods do not fully leverage the information that is shared among cell lines and drugs with similar structure. As such, drug similarities in terms of cell line responses and chemical structures could prove to be useful in forming drug representations to improve drug response prediction accuracy. Results: We present two deep learning approaches, BiG-DRP and BiG-DRP+, for drug response prediction. Our models take advantage of the drugs' chemical structure and the underlying relationships of drugs and cell lines through a bipartite graph and a heterogenous graph convolutional network that incorporate sensitive and resistant cell line information in forming drug representations. Evaluation of our methods and other state-of-the-art models in different scenarios show that incorporating this bipartite graph significantly improve the prediction performance. Additionally, genes that contribute significantly to the performance of our models also point to important biological processes and signaling pathways.

MARSY: A multitask deep learning framework for prediction of drug combination synergy scores

“…In particular, other than the learning rate, number of encoder layers and the choice of optimizer had the largest effect, where 2 encoder hidden layers resulted in 1.4% higher median SCC (0.3% higher median PCC) compared to 3 hidden layers and Adamax resulted in 1.1% higher median SCC (0.5% higher median PCC) compared to Adam (Figure 4C and Supplementary Figure S7). In conclusion, learning rate seems to have the most effect (consistent with our experience in other related studies 12 ), but when large learning rates are excluded, the results are not too sensitive to the choice of hyperparameters. However, if computational complexity is not an issue, marginal improvements can be achieved using hyperparameter tuning based on an independent validation set.…”

“…Due to the distinct molecular and clinical characteristics of cancer types, it is necessary to evaluate the response of cancer cells to different treatments and treatment strategies in each cancer type. The curation of molecular profiles of cancer cell lines (CCLs) and their response to monotherapies in large databases such as the Cancer Cell Line Encyclopedia (CCLE) 9 and Genomics of Drug Sensitivity in Cancer (GDSC) 10 initiated the development of various computational models for prediction of single drug response in CCLs 11,12 and patient tumours 13,14 . More recently, large databases of synergy scores of drug combinations (mainly drug-pairs) in CCLs such as DrugComb 15 have been curated based on results of many high-throughput drug screening studies.…”