What it is
Machine learning models offer potential for predicting the transcriptomic effects of perturbations, but they currently struggle to generalize across cellular contexts. The authors introduce State, a machine learning model that predicts perturbation effects while accounting for cellular heterogeneity within and across experiments; it is trained on single-cell gene expression data to predict perturbation effects across sets of cells. State improved discrimination of perturbation effects on large datasets by more than 30% and identified differentially expressed genes across genetic, signaling and chemical perturbations with significantly improved accuracy compared with baselines. Its cell embeddings, trained on observational data from 167 million cells, enable the identification of strong perturbations in cellular contexts where no perturbations were observed during training. The authors also introduce Cell-Eval, a comprehensive framework for evaluating future models, and say State's performance and flexibility set the stage for scaling the development of AI models of cell state.
Why it matters
State improved discrimination of perturbation effects on large datasets by more than 30% and flagged strong perturbations in cell contexts it never saw perturbed. Generalizing across cellular contexts is where models that predict the transcriptomic effects of perturbation currently struggle. State accounts for heterogeneity within and across experiments, identified differentially expressed genes across genetic, signaling and chemical perturbations with significantly better accuracy than baselines, and comes with Cell-Eval, a framework for evaluating future models.
Underlined numbers link to their source. Every metric and quoted figure is listed under Sources and data below.
Filed undervirtual cell, single-cell, perturbation, transformer, Arc Institute
