Original title: Interpretable Modeling of Single-cell perturbation Responses to Novel Drugs Using Cycle Consistence Learning
Authors: Wei Huang, Aichun Zhu, Hui Liu
This article delves into screening cell-active compounds through phenotype-based methods. It introduces a deep learning system that interprets how cells respond to drug influences using transcriptional and proteomic data. Employing an encoder-decoder structure, the model creates a latent space to represent initial cell conditions, assuming the drug effects adhere to linear addition. By imposing cycle consistency restrictions, it ensures that perturbed states link back to the initial ones and vice versa. This framework, combining consistency checks and linear models, generates understandable and transferable representations of drug impacts, predicting cell responses to new drugs. Validated across diverse datasets—bulk transcriptional and proteomic responses, and single-cell transcriptional data—the model outperforms existing methods, exhibiting improved predictive accuracy for understanding cellular reactions to different drugs.
Original article: https://arxiv.org/abs/2311.10315