Advancing Regulatory Genomics With Machine Learning.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Machine learning (ML) models are increasingly used to predict gene expression and chromatin features directly from DNA sequences.
• These predictive models offer powerful tools for advancing regulatory genomics and uncovering biological insights into gene regulation.

Purpose of the Study:

• To review various ML methods for predicting gene expression and chromatin features from DNA sequence.
• To detail strategies for extracting novel gene-regulation hypotheses from these ML models.
• To discuss methods for quantifying the confidence of ML-derived hypotheses for experimental validation.

Main Methods:

• Review of ML techniques including linear models, random forests, kernel methods, and deep learning.
• Analysis of strategies for hypothesis generation from ML model outputs.
• Examination of confidence scoring procedures for ML-based biological insights.

Main Results:

• ML models provide significant advances in regulatory genomics by predicting molecular features from DNA sequence.
• Diverse ML approaches exist, each with specific strengths for hypothesis generation.
• Confidence measures are crucial for validating ML-derived insights but vary across model types.

Conclusions:

• ML models are valuable tools for generating hypotheses in gene regulation research.
• Confidence assessment is essential for prioritizing ML-generated hypotheses for experimental validation.
• The choice of ML model impacts the nature and reliability of extracted biological insights.