Machine Learning2 min reading time
Generalizable perturbation prediction
Nature
Read the full article
Researchers developed 'State,' a machine-learning model using transformer architecture and pretrained embeddings to predict single-cell perturbation responses across diverse biological contexts. Trained on over 267 million single-cell profiles, State improves generalization and heterogeneity modeling in perturbation prediction. The team also introduced 'Cell-Eval,' a comprehensive framework to evaluate such models beyond traditional accuracy metrics.
