Identifying anal and cervical tumorigenesis-associated methylation signaling with machine learning methods

Area of Science:

• Oncology
• Epigenetics
• Bioinformatics

Background:

• Cervical and anal carcinomas are neoplastic diseases with poorly understood initiation and progression mechanisms.
• Aberrant DNA methylation is implicated in tumorigenesis, suggesting its role as a biomarker for cancer staging.

Purpose of the Study:

• To analyze DNA methylation profiles in cervical and anal carcinoma samples across different tumor progression stages.
• To identify methylation features and develop machine learning classifiers for distinguishing cancer stages.

Main Methods:

• Utilized machine learning algorithms to analyze methylation profiles of cervical and anal carcinoma samples.
• Employed advanced feature selection techniques (Boruta, LASSO, LightGBM, MCFS) to identify key methylation markers.
• Developed classification rules to quantify the impact of methylation on tumorigenesis.

Main Results:

• Identified specific methylation probes (e.g., cg01550828, cg27012396) and associated genes (RNF168, HDAC4) linked to cancer progression.
• Confirmed the role of cg27012396/HDAC4 in regulating glycolysis and survival of hypoxic cervical cancer cells.
• Developed effective classifiers for identifying tumor stages with high accuracy.

Conclusions:

• DNA methylation patterns serve as significant biomarkers for staging cervical and anal carcinomas.
• Machine learning effectively identifies and quantifies methylation signals associated with cancer development.
• This research provides qualitative and quantitative insights into methylation's role in tumorigenesis.