Identifying Transcription Factors That Prefer Binding to Methylated DNA Using Reduced G-Gap Dipeptide Composition.

Area of Science:

• Computational biology
• Genomics
• Machine learning in molecular biology

Background:

• Transcription factors (TFs) regulate gene expression and 3D genome structure by binding to DNA.
• DNA methylation traditionally inhibits TF binding, but some TFs can bind methylated DNA.
• Identifying these TFs is crucial for understanding gene regulation, but experimental methods are laborious.

Purpose of the Study:

• To develop computational methods for identifying transcription factors (TFs).
• To develop computational methods for identifying TFs that bind methylated DNA targets (TFPMs).
• To provide efficient alternatives to experimental techniques for TF and TFPM identification.

Main Methods:

• TF identification: Position-specific scoring matrix for data representation and deep convolutional neural networks for modeling.
• TFPM identification: Reduced g-gap dipeptide composition for data representation and support vector machine algorithm for modeling.
• Both methods were evaluated on independent test sets.

Main Results:

• The TF identification method achieved 90.56% sensitivity, 83.96% specificity, and an AUC of 0.9596.
• The TFPM identification method achieved 82.61% sensitivity, 64.86% specificity, and an AUC of 0.8486.
• Performance metrics surpassed those of existing studies on similar problems.

Conclusions:

• The proposed machine learning models effectively identify TFs and TFPMs.
• These computational approaches offer significant improvements in accuracy and efficiency over experimental methods.
• The findings advance the understanding of gene expression regulation, particularly concerning DNA methylation.