Deep learning framework for predicting gene expression in mutants
A deep learning framework that predicts gene expression patterns in mutants using a language model–based approach. Trained on RNA sequencing data, it identifies expression changes driven by promoter sequence variations and supports mechanistic interpretation. The method leverages local non-coding sequence context near transcription start sites, can use single-cell RNA sequencing, and includes explainable AI tools such as feature attribution and in silico mutagenesis.
Overview
A deep learning framework leverages a large language model to predict cell type-specific gene expression patterns by analyzing local non-coding sequences near transcription start sites. The approach can handle multiple promoter sites and directly utilize single-cell RNA sequencing data, enabling prediction of expression changes in mutants. The framework has been tested on multiple human and mouse datasets, demonstrating its ability to capture the effects of sequence variations on gene expression.
Technical specifications
Key Features:
- Uses a large language model to predict gene expression from non-coding sequences
- Processes and trains on single-cell RNA sequencing (scRNA-seq) data
- Tested on Perturb-seq datasets to predict expression changes due to sequence variants
- Incorporates explainable AI techniques, including feature attribution and in silico mutagenesis, to elucidate the role of regulatory elements in expression changes
- Designed to be generalizable across diverse datasets, including those from multiple species, to improve robustness of predictions
Technology readiness level
The technology is currently at TRL 3, indicating experimental proof of concept. The framework has been tested on existing datasets, and future validation steps include training and validation on diverse plant genomes and additional datasets to further prove effectiveness and adaptability.
About University of California, Davis
UC Davis is a comprehensive public research university in Northern California, combining a land-grant heritage with an integrated academic health system and a veterinary teaching hospital. Industry engages through co-located clinical and translational facilities in Sacramento, a campus-adjacent research park, pilot-scale food and bioprocessing facilities, and a statewide extension network for field-to-market deployment. Proximity to the Bay Area and California's Central Valley enables rapid collaboration with startups and established enterprises, while corporate relations staff streamline access to faculty, talent, and core labs. Research is supported by competitive federal funding from agencies such as NIH, NSF, USDA, and DOE. A dedicated technology transfer office and venture support programs provide IP management, licensing, and startup formation pathways for industry partnerships.
delivered to your inbox.