in-CAHOOTTS: A Neural ODE Framework for Dynamic Gene Regulatory Network Inference and Temporal Modeling

in-CAHOOTTS ( gene regulatory network inference for Context Aware Hybrid neural-ODEs on Transcriptional Time-series Systems) is a biophysically-motivated neural ordinary differential equation framework that simultaneously infers gene regulatory networks and predicts gene expression dynamics from single-cell time-series data.

Key Features

• Biophysical Decomposition: Separates gene expression changes into mRNA transcription and degradation components
• Interpretable Architecture: Maintains full biological interpretability while achieving high predictive accuracy
• Long-term Prediction: Enables unprecedented temporal extrapolation (30x beyond training data)
• Transcription Factor Activity Inference: Estimates latent TF activities that drive regulatory responses
• Prior Knowledge Integration: Incorporates known regulatory interactions to guide network inference

Data

The single-cell RNA sequencing data used in this study is available from GEO under accession number GSE242556. The dataset contains 173,361 Saccharomyces cerevisiae cells sampled over 80 minutes during rapamycin treatment.

For detailed data preprocessing steps, see the Methods section of our bioRxiv preprint.