GeneGeoFlow: AI Model Predicts Cell Responses to Perturbations
A novel AI model named GeneGeoFlow has been introduced for forecasting transcriptional responses of single cells to novel genetic changes and drug combinations. This model, outlined in an arXiv paper (2608.06824), tackles a key challenge in virtual cell modeling. Current graph-based approaches typically utilize a uniform network for gene representation and intergene interactions, often treating stable associations as pathways for perturbation responses. GeneGeoFlow enhances this by conditioning a control-anchored residual flow on gene-specific geometry obtained from biological networks, enabling it to learn transcriptional responses tailored to specific interventions. It employs multi-scale spectral coordinates from Gene Ontology and coexpression networks, along with a gene-wise gating mechanism conditioned on perturbations. This model seeks to enhance the precision of cellular response predictions, vital for drug development and personalized medicine. The paper was presented as a cross-type submission on arXiv.
Key facts
- GeneGeoFlow is a new AI model for virtual cell perturbation modeling.
- It predicts single-cell transcriptional responses to genetic perturbations and drug combinations.
- The model uses gene geometry from biological networks.
- It conditions a control-anchored residual flow on gene-wise geometry.
- It derives multi-scale spectral coordinates from Gene Ontology and coexpression networks.
- The paper is available on arXiv with ID 2608.06824.
- The model aims to learn intervention-specific transcriptional responses.
- It addresses limitations of existing graph-based models.
Entities
Institutions
- arXiv