A Deep Learning Framework for Chromatin Loop De Novo Prediction With Enhanced Feature Extraction.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Chromatin loops are critical 3D structures involved in gene regulation and cellular homeostasis.
• Existing chromatin loop prediction models often fail to account for data heterogeneity, and dedicated tools are scarce.

Purpose of the Study:

• To introduce CHASOS2 (CHromatin loop prediction with Anchor Score and OCR Score), a novel, user-friendly toolkit for de novo prediction and evaluation of chromatin loops.
• To address the limitations of existing models by effectively handling heterogeneous feature data.

Main Methods:

• CHASOS2 utilizes convolutional modules with multi-receptive fields to generate robust features that mitigate data heterogeneity.
• A gradient boosting tree model is employed for predicting chromatin loops based on these generated features.

Main Results:

• Experimental evaluations demonstrate that CHASOS2 outperforms existing prediction methods, especially when dealing with heterogeneous feature data.
• A case study on the K562 cell line showed high consistency between CHASOS2 predictions and experimentally identified chromatin loops (ChIA-PET).

Conclusions:

• CHASOS2 provides an effective and user-friendly solution for de novo chromatin loop prediction.
• The toolkit's ability to handle data heterogeneity and its validated accuracy make it a valuable resource for genomic research.