An Extensive Examination of Discovering 5-Methylcytosine Sites in Genome-Wide DNA Promoters Using Machine Learning...

Area of Science:

• Genomics
• Computational Biology
• Cancer Research

Background:

• Cancer cell proliferation is linked to DNA hypomethylation and promoter hypermethylation of tumor suppressor genes.
• Locating 5-methylcytosine (5mC) sites in promoters is vital for understanding gene expression regulation and cancer development.
• Current experimental methods for genome-wide 5mC identification are time-consuming and labor-intensive.

Purpose of the Study:

• To evaluate the effectiveness of popular machine learning techniques for predicting 5mC sites in genome-wide DNA promoters.
• To apply a novel feature extraction method using k-mer embeddings learned from a language model.
• To compare the predictive performance against existing deep learning models.

Main Methods:

• Utilized XGBoost, Random Forest, Deep Forest, and Deep Feedforward Neural Network models.
• Employed k-mer embeddings derived from a language model for feature extraction.
• Assessed model performance using AUC scores on cross-validation and independent test datasets.

Main Results:

• All machine learning models surveyed outperformed previous deep learning models on the same dataset with a different encoding scheme.
• The best performing model achieved an Area Under the Curve (AUC) score of 0.962.
• The k-mer embedding feature extraction method proved effective in enhancing prediction accuracy.

Conclusions:

• The developed approach is highly efficient for identifying 5mC sites in DNA promoters.
• Machine learning models, particularly when combined with advanced feature extraction, offer a powerful tool for epigenetic research.
• This method provides a high-performance solution for the critical task of genome-wide 5mC site identification.