TFPred

TFPred predicts transcription factors and their binding preferences for methylated versus non-methylated DNA using sequence-derived protein representations and machine learning for applications in epigenetic regulation.

Key Features:

• Two-step discriminative method: Uses a two-step approach to first distinguish transcription factors from non-transcription factors using sequence data.
• Classification of methylation preference: Classifies transcription factors according to preference for methylated versus non-methylated DNA.
• Sequence-based protein representation: Employs sequence information as the primary input and sequence-based methods to represent proteins.
• Machine learning algorithms: Constructs models using Support Vector Machines (SVM) and XGBoost.
• Validation and performance metrics: Reports AUCs of 0.9183 (5-fold CV) and 0.9116 (independent) for TF identification, AUCs of 0.7744 (5-fold CV) and 0.7356 (independent) for methylation-preference classification, and accuracies of 86.66% (5-fold CV) and 83.02% (independent) for TF identification and 71.58% (5-fold CV) and 68.87% (independent) for methylation-preference classification.

Scientific Applications:

• Gene regulation and epigenetics: Identifies transcription factors that preferentially bind methylated DNA to inform studies of epigenetic regulation.
• 3D genome conformation and gene expression: Supports investigation of how DNA methylation influences 3D genome architecture and gene expression.
• Cellular differentiation and methylation-mediated processes: Aids research into methylation-mediated biological processes relevant to cellular differentiation.

Methodology:

Implements a two-step computational pipeline: (Step 1) discriminate transcription factors from non-transcription factors using sequence-based protein representations; (Step 2) classify transcription factors by preference for methylated versus non-methylated DNA using models built with SVM and XGBoost and validated by 5-fold cross-validation and an independent dataset.