Novel AI-powered computational method using tensor decomposition for identification of common optimal bin sizes when...

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Integrating multiple Hi-C datasets requires a common bin size, which is challenging as dataset quality varies with resolution.
• Determining optimal bin sizes is crucial for accurately identifying common structural features in Hi-C data.
• Existing quality assessments have limitations in selecting a single optimal bin size applicable across diverse Hi-C datasets.

Purpose of the Study:

• To propose a novel method for selecting optimal bin sizes for integrating multiple Hi-C datasets.
• To enable automatic empirical estimation of the highest resolution for multi-dataset Hi-C analysis.
• To overcome the limitations of manual threshold setting and dataset-specific quality assessments.

Main Methods:

• Application of tensor decomposition (TD) for unsupervised feature extraction (FE).
• Utilizing phase transition-like phenomena observed in TD-based FE to identify optimal bin sizes.
• Analysis of Hi-C datasets retrieved from GEO (GEO IDs: GSE260760, GSE255264).

Main Results:

• The proposed TD-based unsupervised FE method automatically estimates the smallest possible bin size (highest resolution).
• This approach empirically determines the optimal resolution without requiring tunable parameters.
• Demonstrated ability to optimize bin sizes across multiple Hi-C profiles.

Conclusions:

• Tensor decomposition provides a robust, parameter-free method for optimizing bin sizes in multi-dataset Hi-C integration.
• This technique facilitates more accurate identification of common genomic structures across datasets.
• Represents a novel advancement in analyzing and integrating Hi-C data.