HLA-Binding Peptide Prediction with Structural Docking Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting peptide binding to HLA molecules are limited by biased sampling, low throughput, and failure to consider endogenous processing and transport, leading to inaccurate and inefficient peptide identification.
Innovation Solution
A method using machine learning algorithms trained on structural features from crystal structures of HLA alleles, incorporating protein-peptide docking and molecular dynamics simulations, to predict peptide binding, combining structural and non-structural features for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing neural-network-based algorithms are used for peptide binding prediction, then the prediction can be performed with current methods, but the accuracy is limited due to biased sampling and failure to consider endogenous processing
Solution Approach 1:
The patent transforms the input parameters from simple peptide sequences to three-dimensional structural features including solvent accessibility, secondary structure, and spatial coordinates. This parameter transformation enables the model to capture structural determinants of binding that sequence-based methods miss, directly improving prediction accuracy while maintaining reliability
Solution Approach 2:
The patent introduces protein structure as an intermediary between peptide sequence and binding prediction. By using structural features as mediators, the model indirectly captures endogenous processing effects and transport constraints that are not directly observable from sequence alone, resolving the reliability issue
2Measurement precision
If mass spectrometry-based approaches are used to obtain unbiased peptide profiles, then a large portrait of processed peptides can be obtained, but the cellular input requirement is large which limits throughput
Solution Approach 1:
The patent replaces the physical mass spectrometry measurement system with a computational prediction system based on structural features. This substitution eliminates the need for large cellular inputs and expensive instrumentation, dramatically increasing throughput while maintaining the ability to identify accurate peptide binders through structural analysis
Solution Approach 2:
The patent creates computational models that copy and simulate the binding interactions observed in mass spectrometry studies. By using structural features to generate in-silico predictions, the system reproduces the accurate peptide profiles obtained experimentally without requiring actual cellular material, thus achieving high throughput
3Quantity of substance
If only biochemical affinity measurements of synthetic peptides are used for training, then the training data can be obtained, but biased sampling occurs which skews results and misses subdominant motifs
Solution Approach 1:
The patent changes the training parameters from biochemical affinity values to structural feature-based binding probabilities. This transformation allows the model to learn from diverse structural configurations rather than being biased toward high-affinity synthetic peptides, enabling detection of subdominant motifs and improving overall prediction accuracy
Solution Approach 2:
The patent inverts the traditional approach by not directly measuring binding affinity but instead predicting binding probability from structural features. This inversion allows the model to capture a broader range of binding events including weak and subdominant interactions that are missed in traditional affinity-based training
Data Source
AI summary
The present invention discloses a method for predicting peptides that are capable of binding to HLA molecules that incorporate the crystal structure of HLA molecules. An improved HLA-specific peptide docking workflow is used to simulate the occupancy of a peptide on the binding pocket of an HLA molecule, and three models are trained to predict the binding of the peptide to HLA molecules. The results show that these models predict HLA-allele specific binding peptides with extremely high accuracy.