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

Area of Science:

• Genomics
• Epigenetics
• Computational Biology

Background:

• Chromosomal positioning near nuclear bodies impacts genome functions like transcription.
• Understanding the sequence and epigenomic drivers of genome-wide chromatin spatial organization remains a challenge.

Approach:

• Developed UNADON, a transformer-based deep learning model, to predict the cytological distance of genomic loci to nuclear bodies.
• Utilized both DNA sequence features and epigenomic signals as input for the model.
• Validated UNADON's predictive accuracy across four diverse cell lines (K562, H1, HFFc6, HCT116) and in an unseen cell type.

Key Points:

• UNADON accurately predicts chromatin spatial positioning to nuclear bodies, even when trained on a single cell line.
• The model demonstrates robust performance across different cell types, including unseen ones.
• Identified potential sequence and epigenomic factors governing large-scale chromatin compartmentalization relative to nuclear bodies.

Conclusions:

• UNADON offers novel insights into the relationship between sequence features and large-scale chromatin spatial localization.
• The findings have significant implications for deciphering nuclear structure and its functional consequences.
• This work advances our understanding of genome organization and regulation within the cell nucleus.