Prediction of gene expression using histone modification patterns extracted by Particle Swarm Optimization

Area of Science:

• Genomics
• Epigenetics
• Bioinformatics

Background:

• Histone modifications are crucial for gene transcription regulation.
• Previous machine learning approaches often lacked explainability or focused solely on histone modification abundance.
• The precise role and interaction of various histone modifications in transcription remain an active area of research.

Purpose of the Study:

• To develop and validate a machine learning model that predicts gene expression based on histone modification patterns in gene promoters.
• To investigate whether histone modification patterns are more predictive of gene expression than their abundance.
• To gain insights into the mechanisms of transcription regulation by explaining the model's decision-making process.

Main Methods:

• Trained machine learning models to predict gene expression using histone modification patterns.
• Employed an optimization algorithm to extract predictive histone modification profiles.
• Utilized model explainability techniques to identify key features influencing predictions.

Main Results:

• The PatternChrome algorithm achieved a high Area Under Curve (AUC) score of 0.9029 for binary classification, outperforming previous methods.
• Extracted histone modification patterns demonstrated generalizability across different samples and were largely independent of cellular specificity.
• Model explanations confirmed existing knowledge and uncovered novel aspects of histone modification's role in transcriptional regulation.

Conclusions:

• Histone modification patterns in gene promoters are highly predictive of gene expression.
• The PatternChrome algorithm provides a powerful and explainable tool for studying transcriptional regulation.
• Findings highlight the generalizability of epigenetic patterns in gene regulation.