Inadequacy of Evolutionary Profiles Vis-a-vis Single Sequences in Predicting Transient DNA-Binding Sites in Proteins

Area of Science:

• Computational biology
• Biophysics
• Machine learning

Background:

• Predicting DNA-binding residues is crucial for understanding protein-DNA interactions.
• Current machine learning methods predominantly use Position Specific Substitution Matrices (PSSMs) as features.

Purpose of the Study:

• To evaluate the effectiveness of PSSMs versus single-sequence models for predicting DNA-binding residues.
• To investigate the performance differences in predicting conserved versus transient binding sites.

Main Methods:

• Created a database of highly similar protein pairs with known protein-DNA complexes.
• Compared machine learning models trained on PSSMs with models trained on sparse-encoded single sequences.
• Investigated differential predictability of conserved and transient binding residues.

Main Results:

• PSSMs are effective for conserved binding sites but inadequate for transient ones.
• Single-sequence models outperformed PSSM-based models by up to 8% for transient binding site prediction.
• PSSM-based models are insufficient for predicting high-specificity DNA-binding residues.

Conclusions:

• Evolutionary profiles (PSSMs) have limitations in predicting dynamic DNA-binding residues.
• Single-sequence based models offer superior performance for transient binding site prediction.
• Findings impact the design of specific DNA ligands and protein-DNA complex modeling.