Two sequence- and two structure-based ML models have learned different aspects of protein biochemistry

Area of Science:

• Computational biology
• Protein engineering
• Machine learning in bioinformatics

Background:

• Deep learning models like large language models (LLMs) and 3D Convolutional Neural Networks (CNNs) are increasingly used for predicting protein mutational effects.
• LLMs utilize transformer architectures on protein sequences, while 3D CNNs process voxelized protein structures.

Approach:

• Systematically compared two LLMs against two structure-based CNN models for protein mutation prediction.
• Analyzed prediction accuracy and generalization capabilities across different protein residue types and environments.

Key Points:

• Sequence-based LLMs and structure-based CNNs exhibit distinct prediction strengths and weaknesses.
• Structure-based models excel at predicting buried hydrophobic residues; LLMs are better with solvent-exposed polar/charged residues.
• Prediction accuracies between sequence and structure-based models are largely uncorrelated.

Conclusions:

• A hybrid approach combining predictions from both sequence- and structure-based deep learning models significantly enhances overall prediction accuracy for protein mutational effects.
• Leveraging the complementary strengths of different model architectures offers a promising direction for advancing protein bioinformatics.