Biomedicine · us
Predicting Genetic Interventions from Cell Snapshots: DynaVelo Validates Three Rescue Targets in Mouse B Cells
Which genetic changes can counteract the effects of another mutation? DynaVelo combines single-cell sequencing with dynamic modeling to identify three candidate targets supported by mouse experiments. The validation concerns specific cellular phenotypes; turning these findings into therapies remains some way off.
When one gene loses its function, can adjusting another gene put cells back on a normal course? DynaVelo, an artificial intelligence model developed by researchers at Memorial Sloan Kettering Cancer Center (MSK) and Weill Cornell Medicine, aims to narrow the experimental search for answers to such questions. All three genetic intervention targets proposed by the team received support in mouse B-cell experiments. The study was published in *Nature Biotechnology* on October 21.
The challenge is that single-cell sequencing usually provides only a molecular snapshot at a particular moment, without continuously tracking the same cell. DynaVelo integrates RNA expression and ATAC sequencing data from the same cell, with the latter reflecting which regions of chromatin are more accessible. The model then uses generative neural ordinary differential equations to estimate how gene activity and regulatory states influence one another and simulate the changes that might follow a reduction in gene function.
The researchers chose mouse germinal center B cells as their testing ground. These cells move between the dark zone, where extensive proliferation occurs, and the light zone, where selection takes place. Some leave the germinal center and develop toward memory B cells or plasma cells. The team first built a dynamic model using 7,166 normal mouse cells, then analyzed cells with loss of function in a single allele of Arid1a or Ctcf to investigate how these mutations alter those pathways.
In the model with partial loss of Arid1a function, DynaVelo identified targets including Ctcf, Stat3 and Bcl6 that might counteract the abnormalities. The researchers then tested whether inactivating one allele of a candidate gene against a background of Arid1a loss could improve the imbalance in the ratio of germinal center dark-zone to light-zone cells. Here, “rescue” means restoring a specific cellular phenotype, rather than replacing the originally damaged gene.
The original paper describes the validation process in greater detail than MSK’s research summary: previous research had already provided support for the Ctcf double-heterozygous mice, and the current study used single-cell multiomics to confirm that their dynamics partially returned toward the normal state. The team also established two new mouse models, Arid1a with Bcl6 and Arid1a with Stat3, and used flow cytometry to confirm that the dark-zone to light-zone ratio returned to the level seen in normal mice. All three targets therefore have in vivo animal evidence, although not all molecular changes were fully reversed.
These findings give virtual genetic experiments a testable biological outcome and suggest that the model could help prioritize subsequent experiments. However, cellular trajectories are still inferred from sequencing data, and predictions are affected by data quality and model settings. Genetically reducing gene function also cannot be directly equated with drug treatment. The study has not demonstrated that these interventions can treat human diseases; their applicable contexts, methods of intervention and safety still need to be clarified.