Chemical-induced gene expression profiles provide critical information on the mode of action, off-target effect, and cellar heterogeneity of chemical actions in a biological system, thus offer new opportunities for drug discovery, system pharmacology, and precision medicine. Despite their successful applications in drug repurposing, large-scale analysis that leverages these profiles is limited by sparseness and low throughput of the data. Several methods have been proposed to predict missing values in gene expression data. However, most of them focused on imputation and classification settings which have limited applications to real-world scenarios of drug discovery. Therefore, a new deep learning framework named chemical-induced gene expression ranking (CIGER) is proposed to target a more realistic but more challenging setting in which the model predicts the rankings of genes in the whole gene expression profiles induced by de novo chemicals. The experimental results show that CIGER significantly outperforms existing methods in both ranking and classification metrics for this prediction task. Furthermore, a new drug screening pipeline based on CIGER is proposed to select approved or investigational drugs for the potential treatments of pancreatic cancer. Our predictions have been validated by experiments, thereby showing the effectiveness of CIGER for phenotypic compound screening of precision drug discovery in practice.
A cDNA microarray gene expression database for cancer drug discovery
