D2R2: Discrete Diffusion with Regulation Reinforcement for Single-Cell Perturbation Prediction

A novel computational technique known as D2R2 (Discrete Diffusion with Regulation Reinforcement) has been developed to forecast single-cell transcriptomic reactions to genetic changes. This method, which appears in a preprint on arXiv (2608.15288), redefines perturbation prediction through regulation-guided, gene-wise progressive generation. In contrast to traditional approaches that assess entire expression profiles collectively, D2R2 focuses on the sequence of individual gene responses. It employs a Masked Discrete Diffusion Model to depict expression as ordinal tokens, reconstructing a fully masked profile incrementally, where the responses of generated genes influence those that remain masked. A Regulatory Policy Module initiates the generation policy based on a gene regulatory network derived from control cells, adjusting it according to the perturbation and the current partially generated state. This advancement tackles a significant challenge in functional genomics and virtual-cell modeling, providing a more detailed understanding of cellular responses.