PRIMED: predicting DNA binding residues by leveraging pre-trained protein language models.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Protein-DNA interactions are crucial for fundamental biological processes like gene regulation and genome stability.
• Identifying DNA-binding residues (DBRs) is essential for protein engineering and therapeutic design, but experimental methods are limited by low throughput.
• Computational approaches using protein sequence and structure offer scalable alternatives for DBR prediction.

Purpose of the Study:

• To develop and validate PRIMED, a novel machine learning framework for accurate prediction of DNA-binding residues.
• To integrate diverse protein representations from multiple protein language models for enhanced predictive performance.
• To assess PRIMED's generalizability across various protein datasets with different DBR frequencies.

Main Methods:

• PRIMED utilizes a multilayer perceptron to process concatenated protein representations derived from three distinct protein language models: ESM-2, ESM-3, and ESM-C.
• The framework integrates biochemical and structural properties encoded within these language model representations.
• Predictions were performed on benchmark datasets including Test-46, Test-129, CLAPE-DB, and a newly curated Test-10 K dataset.

Main Results:

• PRIMED achieved high performance on benchmark datasets, with an AUC of 0.92 and MCC of 0.64 on Test-46, and an AUC of 0.93 and MCC of 0.45 on Test-129.
• The model demonstrated strong generalizability on the large-scale Test-10 K dataset, performing competitively against existing methods.
• Results indicate PRIMED's effectiveness in predicting DBRs across proteins with varying DBR percentages.

Conclusions:

• Integrating diverse protein language model representations significantly improves the accuracy and transferability of DNA-binding residue predictions.
• PRIMED offers a powerful computational tool for advancing research in gene regulation, genome stability, and disease mechanisms.
• The framework's performance highlights the potential of advanced machine learning techniques in molecular biology.