Asymmetric Predictive Relationships Across Histone Modifications.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Understanding epigenomic landscapes is crucial for cellular processes and diseases.
• Artificial intelligence offers new epigenome imputation strategies.
• Predictive relationships among epigenetic marks are underexplored.

Purpose of the Study:

• Develop a machine learning approach for epigenomic imputation and interpretation.
• Reveal predictive relationships among six key histone marks.
• Assess the predictive performance of the developed method.

Main Methods:

• Utilized a machine learning framework for epigenomic data imputation.
• Dissected spatial contributions from six histone marks to identify cross-prediction patterns.
• Validated the approach on held-out epigenome data.

Main Results:

• Identified prevalent and asymmetric cross-prediction relationships among histone marks.
• Achieved high predictive performance on prospective epigenomes, surpassing state-of-the-art methods.
• Successfully imputed 527 and 2,455 unavailable genome-wide histone modification signal tracks for ENCODE3 and Roadmap datasets, respectively.

Conclusions:

• The developed machine learning approach enables accurate epigenomic imputation and interpretation.
• Revealed novel insights into the predictive interplay of histone modifications.
• Provides valuable imputed data resources for future epigenomic research.