DNAffinity

DNAffinity predicts transcription factor (TF) binding affinities by combining a physics-based description of DNA structural and mechanical properties derived from atomistic molecular dynamics simulations with machine learning.

Key Features:

• Physics-based machine learning framework: Integrates physical principles of DNA mechanics with machine learning to model TF-DNA interactions.
• Atomistic MD-derived DNA descriptors: Uses structural and mechanical properties of DNA extracted from atomistic molecular dynamics simulations as input features.
• Versatile prediction across assays: Predicts binding affinities for experimental techniques including ultra-sensitive protein-binding microarrays (uPBM), gel-based protein-binding microarrays (gcPBM), and high-throughput SELEX (HT-SELEX).
• Superior predictive performance: Outperforms existing algorithms in predicting TF-DNA binding affinities.
• Extensible to sequence variants and noncanonical bases: Can be adapted to account for epigenetic variants, mismatches, mutations, and non-coding nucleobases.
• Integration with chromatin information: Can incorporate chromatin structure information to extend predictions toward in vivo binding contexts.

Scientific Applications:

• In vitro binding affinity prediction: Provides quantitative TF binding affinity predictions for in vitro assays to support studies of transcriptional regulation.
• In vivo binding site estimation: When combined with chromatin structure information, enables estimation of in vivo TF binding sites, as demonstrated in yeast.

Methodology:

Machine learning models trained on structural and mechanical DNA properties obtained from atomistic molecular dynamics simulations.