Infectious Disease Epitope Prediction With Multi-Part HLA Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The lack of comprehensive mass spectrometry immunopeptidomic datasets for infectious diseases limits the development of accurate models for predicting viral epitopes, hindering the effectiveness of therapeutic vaccines.
Innovation Solution
A multi-part presentation model combining a pan-allele model and allele-specific models to generate per-allele presentation likelihoods for infectious disease-derived antigens, utilizing training data from binding affinity and mass spectrometry data to identify antigens likely to be presented by HLA alleles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If mass spectrometry immunopeptidomic datasets are used for training, then prediction accuracy is improved, but such datasets are lacking for infectious disease-derived antigens
Solution Approach 1:
The patent combines multiple data sources including mass spectrometry immunopeptidomic data, binding affinity data, and sequence-based features into a unified training dataset. This merging approach allows the model to leverage available data from different modalities to compensate for the scarcity of infectious disease-specific immunopeptidomic data, thereby improving prediction accuracy despite limited training data availability.
Solution Approach 2:
The patent develops a universal prediction model that can handle multiple types of input data (mass spectrometry data, binding affinity data, sequence features) and apply to various infectious disease contexts. This multi-functional model architecture allows it to effectively utilize diverse data sources and adapt to different infectious disease scenarios, improving prediction accuracy without requiring large amounts of disease-specific training data for each scenario.
2Measurement precision
If a multi-part presentation model is used, then prediction accuracy is improved, but model complexity increases
Solution Approach 1:
The patent implements a multi-part presentation model that segments the prediction task into distinct components: a mass spectrometry-based model, a binding affinity-based model, and a sequence-based model. Each component processes specific types of data independently and their predictions are integrated to produce the final result. This segmentation improves prediction accuracy by leveraging complementary information from different data sources while maintaining manageable complexity within each individual model component.
Data Source
AI summary
Disclosed herein is a system and methods for determining the alleles, antigens, and infectious disease-based vaccine composition as determined on the basis of a patient's expressed HLA alleles. Additionally described herein are unique infectious disease-derived vaccines.