TxPert: Leveraging BiochemicalRelationships for Out-of-Distribution Transcriptomic Perturbation Prediction

Accurately predicting cellular responses to genetic perturbations is essential for understandingdisease mechanisms and designing effective therapies. Yet exhaustively exploring the spaceof possible perturbations (e.g., multi-gene perturbations or across tissues and cell types) isprohibitively expensive, motivating methods that can generalize to unseen conditions. In thiswork, we explore how knowledge graphs of gene-gene relationships can improve out-of-distribution(OOD) prediction across three challenging settings: unseen single perturbations; unseen doubleperturbations; and unseen cell lines. In particular, we present: (i) TxPert, a new state-of-the-artmethod that leverages multiple biological knowledge networks to predict transcriptional responsesunder OOD scenarios; (ii) an in-depth analysis demonstrating the impact of graphs, modelarchitecture, and data on performance; and (iii) an expanded benchmarking framework thatstrengthens evaluation standards for perturbation modeling.