Modeling nascent transcription from chromatin landscape and structure with CLASTER

Area of Science:

• Genomics
• Epigenetics
• Computational Biology

Background:

• Understanding gene transcription regulation is crucial for deciphering cellular function and disease.
• Current models often struggle to integrate diverse epigenetic data and 3D genome structure.
• Predicting the impact of epigenetic changes on transcription remains a challenge.

Purpose of the Study:

• To introduce CLASTER, an epigenetic-based deep neural network.
• To integrate various data modalities of chromatin landscape and 3D structure.
• To predict nascent transcription levels at kilobasepair resolution.

Main Methods:

• Development of the Chromatin Landscape and Structure to Expression Regressor (CLASTER) deep neural network.
• Integration of multi-modal epigenetic data and 3D chromatin structure information.
• Training the model to predict nascent transcription levels.

Main Results:

• CLASTER successfully translates chromatin landscape and structure data into nascent transcription levels.
• The model achieves kilobasepair resolution in transcription prediction.
• CLASTER serves as a platform for understanding epigenetic drivers of transcription and predicting perturbation effects.

Conclusions:

• The predominant locality in machine learning models reflects genomic organization.
• CLASTER offers insights into the learned rules governing nascent transcription.
• Findings have broad implications for future computational modeling in genomics and epigenetics.