Machine learning algorithms for simultaneous supervised detection of peaks in multiple samples and cell types.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Joint peak detection is crucial for comparing epigenomic samples.
• Existing unsupervised algorithms are limited to two samples and lack interpretability.

Purpose of the Study:

• To introduce PeakSegPipeline, a novel genome-wide, multi-sample peak calling pipeline.
• To enhance accuracy and model interpretability in epigenomic data analysis.

Main Methods:

• Utilizes a constrained maximum likelihood segmentation model.
• Employs a learned penalty function based on user-provided peak/non-peak labels.
• The number of peaks is the primary tunable parameter.

Main Results:

• PeakSegPipeline demonstrates comparable or superior accuracy to state-of-the-art methods.
• The pipeline generates interpretable models with consistently overlapping peaks across samples.
• It effectively learns experiment-specific variations in predicted peak sizes.

Conclusions:

• PeakSegPipeline provides an advanced solution for multi-sample epigenomic peak detection.
• The method offers improved accuracy, interpretability, and adaptability to experimental variations.
• It advances the analysis of large-scale epigenomic datasets.