Introduction

With the continuous advancement of phenotype-based drug discovery research, accurately predicting drug-induced gene expression and achieving efficient compound screening have become critical issues. However, existing methods are inadequate in integrating the effects of time, dose, and cell line on drug response, which greatly limits their effectiveness. In this context, we propose the DeepICER model. The model integrally encodes time and dose with compounds, and employs a bilinear attention mechanism to integrate with the characteristics of cell lines untreated with compounds, and then predicts the differential gene expression profiles of cells perturbed by new chemicals. Here we provide a platform to perform DeepICER predictions online, using data customized by the user to predict the cellular transcriptional changes induced by a drug at a specified time and dose, and to evaluate the effect of the drug based on the given set of signature genes.

Workflow

Cell line