CrossTx: Cross-cell line Transcriptomic Signature Predictions

Abstract

MotivationPredicting the cell response to chemical compounds is central to drug discovery, drug repurposing, and personalized medicine. To this end, large datasets of drug response signatures have been curated, most notably the Connectivity Map (CMap) from the Library of Integrated Network-based Cellular Signatures (LINCS) project. A multitude of in silico approaches have also been formulated to leverage drug signature data for accelerating novel therapeutics. However, the majority of the available data are from immortalized cancer cell lines. Cancer cells display markedly different responses to compounds, not only when compared to normal cells, but also among cancer types. Strategies for predicting drug signatures in unseen cells--cell lines not in the reference datasets--are still lacking. ResultsIn this work we developed a computational strategy, called CrossTx, for predicting drug transcriptomic signatures of an unseen target cell line using drug transcriptome data of reference cell lines and background transcriptome data of the target cells. Our strategy involves the combination of predictor and corrector steps. Briefly, the Predictor applies averaging (mean) or linear regression model to the reference dataset to generate cell line-agnostic drug signatures. The Corrector generates target-specific drug signatures by projecting cell line-agnostic signatures from the Predictor onto the transcriptomic latent space of the target cell line using Principal Component Analysis (PCA) and/or an Autoencoder (AE). We tested different combinations of Predictor-Corrector algorithms in an application to the CMap dataset to demonstrate the performance of our approach. ConclusionCrossTx is an efficacious and generalizable method for predicting drug signatures in an unseen target cell line. Among the combinations tested, we found that the best strategy is to employ Mean as the Predictor and PCA followed by AE (PCA+AE) as the Corrector. Still, the combination of Mean and PCA (without AE) is an attractive strategy because of its computationally efficiency and simplicity, while offering only slightly less accurate drug signature predictions than the best performing combination. Availability and implementationhttp://www.github.com/cabsel/crosstx Contactrgunawan@buffalo.edu