Application of Feature Selection and Deep Learning for Cancer Prediction Using DNA Methylation Markers

Area of Science:

• Genomics
• Bioinformatics
• Machine Learning

Background:

• DNA methylation influences gene accessibility and expression, playing a role in cancer development.
• Accurate breast cancer prediction and identification of contributing genes are crucial for effective treatment.

Purpose of the Study:

• To develop and evaluate a machine learning pipeline for breast cancer prediction using DNA methylation data.
• To identify significant genes and methylation markers associated with breast cancer.
• To enhance prediction accuracy by combining deep learning with feature selection.

Main Methods:

• Utilized The Cancer Genome Atlas (TCGA)-BRCA methylation dataset.
• Applied feature engineering to reduce data volume and enable scalable deep learning.
• Developed a bioinformatics workflow incorporating imputation and data balancing.
• Compared deep learning approaches on Illumina 27K and 450K methylation data.

Main Results:

• Achieved 98.75% prediction accuracy using 450K methylation markers in under 13 seconds.
• Identified significantly enriched biological and molecular functions in reduced gene sets from both 27K and 450K data.
• Discovered seven common oncogene/tumor suppressor genes between the two datasets (RTN4IP1, MYO18B, ANP32A, BRF1, SETBP1, NTRK1, IGF2R).

Conclusions:

• The proposed deep learning workflow accurately identifies key methylation markers and functionally important genes in breast cancer.
• Coupling deep learning with feature selection enhances prediction accuracy compared to standalone methods.
• This approach offers a promising tool for improved breast cancer diagnostics and biomarker discovery.