Motion as a Language: Transformer-Based Classification of Antimicrobial Peptide Conformational Dynamics.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) are vital alternatives to antibiotics due to rising bacterial resistance.
• AMP conformational plasticity is key for target recognition, but data complexity hinders its use in screening.
• Molecular dynamics (MD) simulations generate extensive conformational data that is difficult to analyze and integrate into databases.

Purpose of the Study:

• To apply transformer neural networks to analyze complex conformational data from MD simulations of AMPs.
• To develop an unsupervised classification method for AMP conformational plasticity.
• To integrate conformational dynamics into AMP screening and drug design pipelines.

Main Methods:

• Utilized transformer neural network architecture, commonly used in large language models, to process time-series data of AMP conformations from MD simulations.
• Developed a method for unsupervised classification of AMP conformational plasticity based on learned representations of conformational space.
• Integrated the learned conformational representations with conventional properties for database screening.

Main Results:

• Successfully applied transformer networks to detect temporal and spatial context in AMP conformational dynamics.
• Demonstrated unsupervised classification of AMP conformational plasticity, enabling its use as a screening criterion.
• Showcased the potential of deep learning to incorporate conformational dynamics into drug design.

Conclusions:

• Deep learning, specifically transformer networks, can effectively analyze complex AMP conformational data.
• Unsupervised classification of conformational plasticity enhances AMP screening and drug design.
• This approach restores the importance of conformational dynamics in the development of novel therapeutics.