ESMR4FBP: A pLM-based regression prediction model for specific properties of food-derived peptides optimized multiple...

Area of Science:

• Food Science
• Bioinformatics
• Artificial Intelligence

Background:

• Growing emphasis on food safety drives peptide research in food sources.
• Traditional methods for determining peptide properties are costly and time-consuming.
• Existing artificial intelligence (AI) models for peptide property prediction often lack sufficient accuracy.

Purpose of the Study:

• To develop a novel AI regression model for accurate peptide property prediction.
• To enhance the cost-effectiveness and efficiency of peptide analysis in food science.
• To improve upon the accuracy of existing state-of-the-art (SOTA) peptide prediction models.

Main Methods:

• Integration of the ESM-2 model with the Long Short-Term Memory (LSTM) architecture.
• Optimization of the integrated model structure using three metaheuristic algorithms: Whale Optimization Algorithm (WOA), Sine Cosine Algorithm (SSA), and Harris Hawks Optimization (HHO).
• Validation of the model using antioxidant tripeptide and bitter peptide datasets.

Main Results:

• The developed model achieved an R² of 0.9458 and RMSE of 0.3135 on the antioxidant tripeptide dataset.
• Performance on the antioxidant tripeptide dataset surpassed the SOTA model by 11.66% in R² and 50.00% in RMSE.
• On the bitter peptide dataset, the model yielded an R² of 0.8385 and RMSE of 0.4414.

Conclusions:

• The developed AI model demonstrates significant potential for accurately predicting various peptide properties.
• The integration of ESM-2, LSTM, and metaheuristic optimization offers a powerful approach for peptide analysis.
• This advancement can contribute to more efficient and cost-effective food safety assessments and peptide research.