SLIM: A small linear model with STRING embeddings for single-cell genetic perturbation prediction

This bioRxiv preprint introduces SLIM as a minimal extension of bilinear models, leveraging STRING embeddings to incorporate prior knowledge of protein interactions. The authors benchmark it against state-of-the-art deep learning methods and show that simple baselines often match or outperform complex models, highlighting the importance of informative priors over architecture complexity. The model is designed for tasks like predicting perturbation outcomes across genes and cell types, which could aid in prioritizing therapeutic targets. As a preprint, it has not yet undergone peer review, but its findings could steer future work toward more interpretable and efficient models in single-cell perturbation analysis.