meth-SemiCancer: a cancer subtype classification framework via semi-supervised learning utilizing DNA methylation...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate cancer subtype identification is critical for patient diagnosis and treatment.
• DNA methylation signatures are key factors in tumorigenesis and hold potential as cancer subtype markers.
• Existing methods struggle with high-dimensional DNA methylome data and limited labeled samples.

Purpose of the Study:

• To develop a semi-supervised cancer subtype classification framework using DNA methylation profiles.
• To address the challenge of limited labeled data in methylome datasets for cancer classification.

Main Methods:

• Developed meth-SemiCancer, a semi-supervised learning framework.
• Pre-trained the model on labeled DNA methylation datasets.
• Generated pseudo-subtypes for unlabeled data.
• Fine-tuned the model using both labeled and unlabeled datasets.

Main Results:

• meth-SemiCancer achieved superior performance compared to standard machine learning classifiers.
• The framework demonstrated higher average F1-score and Matthews correlation coefficient.
• Fine-tuning with pseudo-labeled data improved generalization compared to supervised methods.

Conclusions:

• meth-SemiCancer offers an effective approach for cancer subtype classification using DNA methylation data.
• The semi-supervised strategy enhances model generalization by leveraging unlabeled samples.
• The framework is publicly available for broader research use.