UNADON: transformer-based model to predict genome-wide chromosome spatial position

Area of Science:

• Genomics
• Molecular Biology
• Computational Biology

Background:

• Chromatin spatial positioning near nuclear bodies influences genome functions like transcription.
• Understanding the sequence and epigenomic factors governing this positioning is crucial but limited.

Purpose of the Study:

• To develop a deep learning model for predicting genome-wide chromatin spatial positioning relative to nuclear bodies.
• To identify sequence and epigenomic determinants of large-scale chromatin organization.

Main Methods:

• Developed UNADON, a transformer-based deep learning model.
• Utilized sequence features and epigenomic signals as input.
• Employed TSA-seq data for measuring cytological distance to nuclear bodies.

Main Results:

• UNADON accurately predicts chromatin spatial positioning across multiple cell lines, even in unseen cell types.
• The model demonstrates high predictive accuracy when trained on a single cell line.
• Identified potential sequence and epigenomic factors influencing chromatin compartmentalization within nuclear bodies.

Conclusions:

• UNADON offers novel insights into the relationship between sequence features and large-scale chromatin spatial localization.
• The findings advance the understanding of nuclear structure and function.
• The source code is publicly available for further research.