LigandRFs: random forest ensemble to identify ligand-binding residues from sequence information alone

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biochemistry

Background:

• Protein-ligand binding is crucial for protein function, with binding sites being key residues.
• Current prediction methods often rely on known protein structures, which are not always available.
• There is a need for sequence-based approaches to predict protein-ligand binding sites.

Purpose of the Study:

• To develop a sequence-based computational method for identifying protein-ligand binding residues.
• To address the challenge of limited structural information in protein binding site prediction.
• To improve the accuracy and accessibility of protein-ligand binding site prediction.

Main Methods:

• A novel sequence-based approach was developed, utilizing feature vectors derived from protein sequences.
• A combination technique was employed to mitigate the impact of varying sliding residue windows during feature encoding.
• To address data imbalance, balanced datasets were created, and a Random Forest (RF)-based classifier was trained on each.
• An ensemble of RF classifiers was constructed to form the final predictor.

Main Results:

• The proposed method effectively identifies protein-ligand binding residues using only sequence information.
• The combination technique and balanced datasets helped improve prediction robustness.
• The ensemble of Random Forest classifiers achieved high predictive accuracy.

Conclusions:

• The developed sequence-based method provides a viable alternative to structure-dependent approaches for predicting protein-ligand binding sites.
• Experimental results on CASP9 and CASP8 datasets show favorable comparisons with state-of-the-art methods.
• This approach enhances the ability to predict binding sites when structural data is scarce.