Beyond Independent Genes: Learning Module-Inductive Representations for Single-Cell Gene Perturbation Prediction

Predicting transcriptional responses to genetic perturbations is a central problem in functional genomics. In practice, perturbation responses are rarely gene-independent but instead manifest as coordinated, program-level transcriptional changes among functionally related genes. However, most existing methods do not explicitly model such coordination, due to gene-wise modeling paradigms and reliance on static biological priors that cannot capture dynamic program reorganization. To address these limitations, we propose scBIG, a module-inductive perturbation prediction framework that explicitly models coordinated gene programs. scBIG induces coherent gene programs from data via Gene-Relation Clustering, captures inter-program interactions through a Gene-Cluster-Aware Encoder, and preserves modular coordination using structure-aware alignment objectives. These structured representations are then modeled using conditional flow matching to enable flexible and generalizable perturbation prediction. Extensive experiments on multiple single-cell perturbation benchmarks show that scBIG consistently outperforms state-of-the-art methods, particularly on unseen and combinatorial perturbation settings, achieving an average improvement of 6.7% over the strongest baselines. The code is available at https://github.com/ttruan2426-dot/scBIG.

Cells respond to genetic changes in coordinated ways: changing one gene often affects whole groups of related genes, much like adjusting one instrument can change the sound of an orchestra. Predicting these responses is important for understanding how genes control cell behavior, which can help researchers study diseases and identify possible treatment targets. However, many existing machine learning methods treat genes too independently, or rely on fixed biological knowledge that may not reflect how gene groups reorganize under different perturbations. We introduce scBIG, a machine learning method that predicts how cells respond when genes are perturbed. Instead of modeling each gene separately, scBIG learns groups of genes that tend to act together, studies how these groups influence one another, and preserves this group-level structure when making predictions. This allows the model to better capture the coordinated nature of cellular responses. Across several single-cell perturbation datasets, scBIG outperformed strong existing methods, especially when predicting responses to new or combined perturbations that were not seen during training. Our results suggest that explicitly modeling gene programs can make perturbation prediction more accurate and more useful for biological discovery.