TempoMAGE: a deep learning framework that exploits the causal dependency between time-series data to predict histone...

Area of Science:

• Epigenetics and Genomics
• Computational Biology
• Developmental Biology

Background:

• Histone tail modifications are crucial for understanding development and disease, often studied using ChIP-seq in time-series experiments.
• Current ChIP-seq methods capture discrete time points, leading to information gaps in early or intermediate developmental stages.
• Existing machine learning models for histone mark prediction do not leverage the temporal dependencies inherent in time-series data.

Purpose of the Study:

• To develop a deep learning model, TempoMAGE, that predicts histone modifications in time-series data by integrating sequence, gene expression, and chromatin accessibility.
• To address the challenge of missing data in time-series experiments, particularly for early developmental stages where material is limited.
• To improve the prediction of enhancer activity and functional annotation using temporal epigenetic information.

Main Methods:

• Trained a deep learning model (TempoMAGE) to predict H3K27ac presence/absence in open chromatin regions.
• Integrated sequence data, gene expression, chromatin accessibility, and temporal changes in H3K27ac state from a reference time-point.
• Evaluated model performance by assessing the impact of reference time-point information and analyzing learned sequence signatures.

Main Results:

• Incorporating reference time-point information consistently enhanced the model's predictive performance.
• Sequence signatures learned by TempoMAGE were specific to the training data, indicating effective feature extraction.
• TempoMAGE successfully predicted enhancer activity from an in vivo dataset, demonstrating its utility for functional annotation.

Conclusions:

• TempoMAGE effectively predicts histone modifications in time-series experiments by leveraging temporal dependencies.
• The model's ability to learn data-specific features and predict enhancer activity highlights its potential for epigenetic research.
• TempoMAGE offers a valuable tool for functional annotation and understanding dynamic epigenetic processes in development and disease.