Network-based Biased Tree Ensembles (NetBiTE) for Drug Sensitivity Prediction and Drug Sensitivity Biomarker Identification in Cancer

Abstract

We present the Network-based Biased Tree Ensembles (NetBiTE) method for drug sensitivity prediction and drug sensitivity biomarker identification in cancer using a combination of prior knowledge and gene expression data. Our devised method consists of a biased tree ensemble that is built according to a probabilistic bias weight distribution. The bias weight distribution is obtained from the assignment of high weights to the drug targets and propagating the assigned weights over a protein-protein interaction network such as STRING. The propagation of weights, defines neighborhoods of influence around the drug targets and as such simulates the spread of perturbations within the cell, following drug administration. Using a synthetic dataset, we showcase how application of biased tree ensembles (BiTE) results in significant accuracy gains at a much lower computational cost compared to the unbiased random forests (RF) algorithm. We then apply NetBiTE to the Genomics of Drug Sensitivity in Cancer (GDSC) dataset and demonstrate that NetBiTE outperforms RF in predicting IC50 drug sensitivity, only for drugs that target membrane receptor pathways (MRPs): RTK, EGFR and IGFR signaling pathways. We propose based on the NetBiTE results, that for drugs that inhibit MRPs, the expression of target genes prior to drug administration is a biomarker for IC50 drug sensitivity following drug administration. We further verify and reinforce this proposition through control studies on, PI3K/MTOR signaling pathway inhibitors, a drug category that does not target MRPs, and through assignment of dummy targets to MRP inhibiting drugs and investigating the variation in NetBiTE accuracy.

Toward Explainable Anticancer Compound Sensitivity Prediction via Multimodal Attention-Based Convolutional Encoders

“…Each of the 985 cell lines was initially represented by the expression levels of 17,737 genes which we then reduced to a subset of 2128 genes through network propagation over the STRING protein-protein interaction (PPI) network (Szklarczyk et al, 2014), a comprehensive PPI database including interactions from multiple data sources. Following the procedure described in Oskooei et al (2018b), STRING was used to incorporate intracellular interactions in our model by adopting a network propagation scheme for each drug, where the weights associated with each of the reported targets were diffused over the STRING network (including interactions from all the evidence types) leading to an importance distribution over the genes (i.e., the vertices of the network). Our adopted weighting and network propagation scheme consisted of the following steps: we first assigned a high weight (W = 1) to the reported drug target genes while assigning a very small positive weight (ε = 1e−5) to all other genes.…”

In Silico Inference of Synthetic Cytotoxic Interactions from Paclitaxel Responses

“…Several studies have used machine learning to predict the response to cancer drugs [ 29 , 30 , 31 , 32 , 33 ]. This advanced predictive model successfully predicts drug responses based on genomic data.…”

Computational Model Reveals a Stochastic Mechanism behind Germinal Center Clonal Bursts

“…An alternative approach is to focus on predicting binding affinity directly from sequence and transcriptomic data. This is already routinely performed to predict drug sensitivity, where the binding affinity between a drug compound and a protein is predicted using transcriptomic data and chemical information without explicitly accounting for the 3D structure of either molecule [77,78]. An advantage of such approaches is that they enable the in silico engineering of new compounds with improved biochemical properties [79].…”