Poster #77: Jeffery Lin

ContrastLEMBAS: Generalizable Drug-Target Prediction via Contrastive Learning

Jeffery Lin¹, Nikolaos Meimetis, PhD², Douglas A. Lauffenburger, PhD² ¹Department of Biology and the Department of Electrical Engineering and Computer Science (EECS), Massachusetts Institute of Technology ²Department of Biological Engineering, Massachusetts Institute of Technology

Predicting how a cell’s gene expression changes in response to a drug is crucial for understanding its mechanism of action (MoA), identifying potential off-target side effects, and accelerating drug discovery. However, many models struggle to generalize to drugs unseen during training, especially when those drugs are chemically dissimilar to those in the training set. LEMBAS, a mechanistic and biology-informed recurrent neural network, has been shown to effectively model intracellular signaling networks to predict transcriptional responses. Additional work has extended LEMBAS to drug transcriptional responses by inferring drug-target interactions, but it relies on a fixed drug-target space, limiting its ability to generalize to drugs or protein targets outside that space.