Substituted Oligosaccharides as Protein Mimics: Deep Learning Free Energy Landscapes.

Area of Science:

• Carbohydrate chemistry
• Computational chemistry
• Biophysics

Background:

• Protein-protein interactions are crucial for cellular functions.
• Designing synthetic mimics of protein interfaces is challenging.
• Oligosaccharides offer potential as protein mimics due to favorable ADMET properties, but their conformational dynamics are less understood than polypeptides.

Purpose of the Study:

• To investigate the conformational landscapes of substituted glucopyranose oligomers as protein interface mimics.
• To develop deep learning models for predicting oligosaccharide stability and generating novel mimic structures.
• To gain insights into the factors governing oligosaccharide conformational dynamics.

Main Methods:

• Microsecond-time-scale enhanced-sampling molecular dynamics simulations were performed on 956 substituted glucopyranose oligomers (length 3-12).
• Deep convolutional neural networks (CNNs) were trained on simulation data to predict stability of longer oligosaccharides from trimer motifs.
• Deep generative adversarial networks (GANs) were developed to propose conformations for arbitrary length oligosaccharide mimics.

Main Results:

• The conformational landscapes of a large set of oligosaccharide mimics were mapped.
• CNNs demonstrated predictive power for oligosaccharide stability based on substructure analysis.
• GANs successfully generated plausible conformations for diverse oligosaccharide mimic sequences and lengths, suitable for docking simulations.

Conclusions:

• Oligosaccharides present a viable, yet underexplored, class of molecules for protein mimic design.
• AI-driven approaches, including CNNs and GANs, are effective for characterizing and designing complex carbohydrate structures.
• Understanding collective effects in oligosaccharide dynamics is key to optimizing their function as biomimetic agents.