Researchers have developed a new approach called $D^{2}R^{2}$ to predict how individual genes respond to genetic perturbations in single cells. Unlike previous methods that predict entire expression profiles at once, $D^{2}R^{2}$ generates gene responses one by one, allowing for better modeling of the order in which they occur. The approach uses a combination of machine learning and regulatory biology to generate gene responses that are consistent with biological prior knowledge. Experiments show that $D^{2}R^{2}$ outperforms previous methods on several metrics and is more reliable than uncertainty-based heuristics. --- Why it matters: $D^{2}R^{2}$ matters because it provides a new way to model the complex interactions between genes in single cells, which can help researchers better understand how genetic perturbations affect cellular behavior. This has implications for fields such as functional genomics and virtual-cell modeling. Source: https://arxiv.org/abs/2608.15288