DNAshapedTFBS

DNAshapedTFBS predicts transcription factor binding sites in ChIP-seq datasets by integrating DNA sequence encodings with DNA shape features to model TF-DNA interactions.

Key Features:

• Integration of Sequence and Shape Information: Combines position weight matrices (PWMs), ternary frequency-formatted matrices (TFFMs), or 4-bit binary encodings with DNA shape features—helix twist, minor groove width, propeller twist, and roll—sourced from the GBshape browser.
• Machine Learning Approach: Applies gradient boosting classifiers to learn complex patterns from combined sequence and DNA shape features for TFBS prediction.
• Versatility Across Models: Supports multiple nucleotide encoding schemes and demonstrates improved predictive power when augmenting position-specific scoring matrix (PSSM) scores with DNA shape data.

Scientific Applications:

• Enhanced Predictive Accuracy: Evaluated on 400 human ChIP-seq datasets covering 76 transcription factors, showing improved TFBS prediction accuracy when including DNA shape features.
• Specificity to Transcription Factor Families: Particularly beneficial for E2F and MADS-domain transcription factor families where DNA shape contributes to binding specificity.

Methodology:

Sequence data are encoded using PWMs, TFFMs, or 4-bit binary representations and combined with DNA structural features from the GBshape browser; gradient boosting classifiers are trained on this integrated dataset to predict TFBSs.