OCRFinder: a noise-tolerance machine learning method for accurately estimating open chromatin regions

Area of Science:

• Genomics
• Epigenetics
• Computational Biology

Background:

• Open chromatin regions (OCRs) are crucial for cellular functions and gene regulation.
• Chromatin accessibility influences gene expression and cellular activities.
• Current methods like ATAC-seq and cfDNA-seq detect OCRs, with cfDNA-seq offering broader biomarker potential.

Purpose of the Study:

• To develop a robust computational method for estimating open chromatin regions from cfDNA-seq data.
• To address the challenge of noisy training data in cfDNA-seq analysis due to variable chromatin accessibility.
• To improve the accuracy and sensitivity of OCR detection.

Main Methods:

• Proposed OCRFinder, a learning-based approach incorporating ensemble learning and semi-supervised strategies.
• Designed OCRFinder with a noise-tolerance mechanism to handle imperfect training labels.
• Evaluated OCRFinder against various noise control strategies and state-of-the-art methods.

Main Results:

• OCRFinder demonstrated superior accuracy and sensitivity in identifying open chromatin regions compared to existing approaches.
• The method effectively mitigated issues arising from noisy labels in cfDNA-seq data.
• OCRFinder also showed strong performance in comparative experiments using ATAC-seq and DNase-seq data.

Conclusions:

• OCRFinder provides a reliable and efficient computational solution for OCR estimation from cfDNA-seq.
• The noise-tolerant design makes it suitable for real-world biological data with inherent variability.
• This advancement facilitates more effective genomic and epigenetic research through improved OCR identification.