Cyclic Peptide Structural Ensemble Prediction Without Full MD Runtime

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing computational methods struggle to predict the complete structural ensembles of cyclic peptides, particularly those adopting multiple conformations, which are crucial for understanding their biological properties and functions.

Innovation Solution

A method combining molecular dynamics simulation results with machine learning to train models that can rapidly predict the structural ensembles of cyclic peptides, using weight vectors and neural networks to determine population distributions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If molecular dynamics simulation is used to predict cyclic peptide structures, then prediction accuracy is improved, but computational time increases significantly

Engineering Contradiction:

Improvestructure prediction accuracyVSAvoidcomputational time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs molecular dynamics simulations in advance to generate training data for machine learning models. By pre-computing structural ensembles for various cyclic peptides and storing them as training datasets, the system enables rapid prediction of new peptide structures without requiring time-consuming real-time simulations during the prediction phase.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates machine learning models that copy and generalize the patterns learned from molecular dynamics simulation data. The models capture the essential structural relationships and conformational preferences observed in MD simulations, allowing them to predict structures of new cyclic peptides by applying learned patterns rather than performing exhaustive simulations.

Inventive Principle:

Principle #26Copying

2Measurement precision

If traditional computational methods are used for cyclic peptide structure prediction, then computational resources are consumed, but complete structural ensembles cannot be accurately predicted

Engineering Contradiction:

Improvestructural ensemble prediction accuracyVSAvoidcomputational method complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex task of predicting complete structural ensembles into multiple components: (1) generating representative conformations through molecular dynamics, (2) clustering similar conformations into structural ensembles, (3) training machine learning models on the segmented data, and (4) using the trained models to predict ensembles for new peptides. This segmentation makes the overall problem more manageable and solvable.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent changes the approach from direct computational prediction to a two-stage process: first generating training data through MD simulations with specific parameters (force fields, temperature, solvent models), then using machine learning models trained on this data to make predictions. This parameter change transforms the prediction problem into a data-driven approach that can handle complex structural ensembles more effectively.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250372195A1Cyclic peptide structure prediction via structural ensembles achieved by molecular dynamics and machine learning

Publication Date: 2025.12.04 TRUSTEES OF TUFTS COLLEGE

AI summary

Disclosed herein are methods and systems for using molecular dynamics simulation results as training datasets for machine-learning models that can provide predictions of cyclic peptide structural ensembles.