DoFormer
DoFormer is a Transformer for single-cell genomics that encodes gene interventions within attention, distinguishing observational and interventional regimes without requiring a directed acyclic graph. Using biologically informed training objectives, DoFormer improves prediction of held-out gene perturbations relative to baseline methods and transcriptomic foundation models. Preprint on bioRxiv.
Alireza Karbalayghareh
Senior Research Scientist, AI/ML
My research focuses on developing universal AI models of gene regulation, capturing cis-regulation, trans-regulation, and intercellular signaling by integrating multimodal, multiomic observational, spatial, time-resolved, and perturbational data from immune cells across diverse biological contexts to understand and engineer the immune system to prevent and treat disease.