Predicting the effect of CRISPR-Cas9-based epigenome editing.

Area of Science:

• Genomics and Epigenetics
• Computational Biology
• Molecular Biology

Background:

• Epigenetic regulation controls mammalian transcription, but direct functional links are not fully understood.
• Histone post-translational modifications (PTMs) are key epigenetic marks influencing gene expression.

Purpose of the Study:

• To develop machine learning models predicting gene expression from histone PTMs using ENCODE data.
• To experimentally validate model predictions, particularly the role of H3K27ac, using epigenome editing tools.
• To assess model performance in predicting expression changes from endogenous signatures versus experimental perturbations.

Main Methods:

• Trained machine learning models on epigenomic and transcriptomic data from 13 human cell types.
• Utilized MNase-seq for genome-wide nucleosome occupancy mapping in HEK293T cells.
• Employed the dCas9-p300 system for targeted histone acetylation (H3K27ac) in HEK293T and K562 cells.

Main Results:

• Models achieved high correlations (∼0.70-0.79) predicting gene expression from histone PTMs across cell types.
• Validated known associations, such as H3K27ac near the transcription start site enhancing expression.
• Models accurately ranked relative gene expression changes after epigenome editing but showed reduced accuracy for within-gene fold-changes compared to cell-type predictions.

Conclusions:

• Machine learning effectively predicts gene expression from endogenous epigenetic states.
• Predicting expression changes from epigenome editing perturbations is more challenging, indicating a need for improved datasets and causal models.
• Advancements in epigenome editing datasets and modeling are crucial for understanding and controlling the human epigenome for health applications.