fourSynergy: ensemble-based interaction calling on 4C-seq data using gradient-free optimization

Area of Science:

• Genomics and Molecular Biology
• Bioinformatics and Computational Biology

Background:

• Chromatin organization is vital for gene regulation and implicated in diseases like cancer.
• Understanding reversible chromatin alterations can lead to novel therapeutic development.
• Circular Chromosome Conformation Capture Sequencing (4C-seq) identifies interactions between genes and regulatory elements.

Purpose of the Study:

• To develop an ensemble algorithm for enhanced 4C-seq chromatin interaction calling.
• To leverage synergies among existing 4C-seq analysis tools for improved accuracy.
• To create an accessible framework for 4C-seq data analysis.

Main Methods:

• Employed an ensemble approach using a weighted-voting strategy with existing 4C-seq algorithms.
• Optimized tool weights using gradient-free optimization based on performance metrics.
• Validated the ensemble approach using leave-one-group-out cross-validation.

Main Results:

• The ensemble method significantly improved predictive performance for chromatin interaction detection.
• Achieved a mean F1-score of 0.31 and mean AUPRC of 0.34, outperforming individual tools (0.13 F1, 0.16 AUPRC).
• Integrated the approach into fourSynergy, a user-friendly 4C-seq analysis framework with a Snakemake pipeline, R/Bioconductor package, and Shiny application.

Conclusions:

• Ensemble approaches enhance predictive performance in 4C-seq chromatin interaction detection compared to individual algorithms.
• This work provides curated 4C-seq datasets and an improved analysis framework.
• The findings support the potential of ensemble methods for advancing chromatin interaction analysis and therapeutic development.