scHiCyclePred: a deep learning framework for predicting cell cycle phases from single-cell Hi-C data using...

Area of Science:

• Genomics
• Molecular Biology
• Computational Biology

Background:

• Single-cell Hi-C (scHi-C) technology enables studying 3D chromatin structure across the cell cycle.
• Accurate prediction of cell cycle phases from scHi-C data is challenging.

Purpose of the Study:

• To introduce scHiCyclePred, a novel prediction model for cell cycle phase determination using scHi-C data.
• To leverage multi-scale 3D chromatin interaction features for improved prediction accuracy.

Main Methods:

• scHiCyclePred integrates multiple feature sets, including multi-scale interaction information, to capture 3D chromatin structure.
• Comparative analysis against Nagano_method and CIRCLET was performed using standard metrics (ACC, F1, Precision, Recall, BACC).
• Evaluation was conducted on a complex tissue dataset (Liu_dataset).

Main Results:

• scHiCyclePred significantly outperforms existing methods like Nagano_method and CIRCLET in predicting cell cycle phases.
• On the Liu_dataset, scHiCyclePred showed substantial improvements over CIRCLET: +0.39 (ACC), +0.52 (F1), +0.52 (Precision), +0.39 (Recall).
• The study provides new insights into 3D chromatin dynamics and gene features during the cell cycle.

Conclusions:

• scHiCyclePred offers a robust method for predicting cell cycle phases from scHi-C data.
• The model enhances the understanding of cell cycle-dependent chromatin organization.
• Findings have implications for cell biology and disease research, with the tool available on GitHub.