Prediction of condition-specific regulatory genes using machine learning.

Area of Science:

• Genomics
• Computational Biology
• Plant Science

Background:

• Genomic technologies generate vast data on protein-DNA interactions and open chromatin.
• Identifying condition-specific transcription factor functions from this data is a major challenge.

Purpose of the Study:

• To develop a novel computational method, ConSReg, for integrating regulatory genomic data into predictive machine learning models.
• To identify key regulatory genes, specifically transcription factors, controlling gene expression in response to various conditions.

Main Methods:

• Developed ConSReg, a machine learning approach integrating diverse genomic datasets.
• Applied ConSReg to Arabidopsis thaliana datasets including single-cell gene expression and abiotic stress treatments.
• Validated ConSReg using independent plant nitrogen response datasets and single-cell RNA-seq data.

Main Results:

• ConSReg accurately predicted transcription factors regulating differentially expressed genes with an average auROC of 0.84.
• Performance was 23.5-25% better than traditional enrichment-based approaches.
• ConSReg outperformed other plant tools in ranking correct transcription factors by three times in independent datasets.
• Successfully identified candidate regulatory genes controlling cell wall formation in Arabidopsis single-cell RNA-seq data.

Conclusions:

• ConSReg offers a robust and accurate method for identifying condition-specific regulatory genes using integrated genomic data.
• The approach significantly improves upon existing methods for transcription factor prediction in plants.
• ConSReg provides a valuable new tool for dissecting gene regulatory networks in eukaryotic systems.