TempoMAGE

TempoMAGE predicts H3K27ac histone modification states within open chromatin regions across time-series datasets to infer temporal chromatin dynamics when ChIP-seq data are missing at intermediate time-points.

Key Features:

• Causal Dependency Exploitation: Leverages causal dependencies in time-series data to predict histone marks at unobserved time-points.
• Integration of Multi-Modal Data: Integrates sequence data, gene expression profiles, chromatin accessibility metrics, and estimated changes in H3K27ac state from a reference time-point.
• Improvement with Reference Time-Point Information: Incorporates information from a reference time-point to systematically improve predictive performance for H3K27ac.
• Data-Specific Feature Learning: Extracts sequence signatures unique to the training dataset to learn data-specific features for accurate prediction.

Scientific Applications:

• Functional Annotation of Enhancers: Predicts enhancer activity using pre-validated in-vivo datasets to support functional annotation of putative enhancers.
• Time-Series Experimental Analysis: Fills gaps in time-series experiments by extrapolating H3K27ac states at stages lacking ChIP-seq measurements.

Methodology:

Employs deep learning trained on sequence information, gene expression levels, chromatin accessibility, and estimated changes in H3K27ac from a reference time-point, leveraging causal time-series dependencies and learned sequence signatures to predict presence or absence of H3K27ac at unobserved time-points.