HLA Class II Epitope Prediction Using Mass Spectrometry-Trained AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting HLA class II-specific epitopes and characterizing CD4+ T cells are not accurate, leading to unclear pathways for tumor-specific antigen presentation and low therapeutic efficacy, with existing predictors like NetMHCIIpan having low throughput and requiring radioactive reagents, and lacking consideration of processing rules.
Innovation Solution
A method using machine learning models trained with mass spectrometry data to predict HLA class II peptide presentation and binding, incorporating quality metrics to improve accuracy, and utilizing biological variables such as gene expression, cleavability, and cellular localization to identify allele-specific epitopes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If mass spectrometry data is used to train machine learning models for HLA class II epitope prediction, then measurement precision and positive predictive value are improved, but device complexity and data processing requirements increase
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models with mass spectrometry data before actual epitope prediction. The models are trained in advance on large datasets of known HLA-peptide interactions, processing rules, and antigen presentation pathways. This pre-computed knowledge base enables rapid and accurate prediction without requiring complex real-time analysis during therapeutic development
Solution Approach 2:
The patent uses machine learning models as intermediaries between raw mass spectrometry data and epitope prediction results. These models serve as mediators that translate complex spectral data into actionable predictions about which peptides will be presented by specific HLA class II molecules. The intermediary models handle the computational complexity while providing simplified, accurate output for therapeutic design
2Ease of operation
If existing predictors like NetMHCIIpan are used for HLA class II epitope prediction, then ease of operation is maintained, but measurement precision and throughput are reduced
Solution Approach 1:
The patent applies parameter changes by modifying the input parameters and training data characteristics of prediction models. Instead of using standard affinity measurements alone, the models incorporate mass spectrometry-derived parameters including actual peptide presentation data, processing rules for different antigen presentation pathways (autophagy, phagocytosis, proteasomal), and allele-specific binding preferences. These parameter changes dramatically improve prediction accuracy while maintaining computational accessibility
3Ease of operation
If affinity measurements are used as the field standard for HLA class II prediction, then ease of operation is maintained, but productivity and measurement precision are reduced due to low throughput and radioactive reagent requirements
Solution Approach 1:
The patent replaces mechanical/chemical affinity measurement systems with computational machine learning models. Instead of performing physical binding assays with radioactive reagents, the system uses in silico predictions based on mass spectrometry training data. This substitution eliminates the need for low-throughput wet lab experiments while dramatically increasing productivity and eliminating radioactive material handling
Solution Approach 2:
The patent creates computational copies of affinity measurement data through mass spectrometry-based proteomics. Rather than performing actual binding assays, the system uses MS data to generate virtual representations of peptide-HLA interactions. These computational copies capture the essential binding information without requiring physical reagents or low-throughput experimental procedures
4Device complexity
If existing methods are used for identifying HLA class II epitopes, then device complexity is minimized, but measurement precision is reduced due to lack of processing rules consideration and allele-specific accuracy
Solution Approach 1:
The patent applies segmentation by dividing the epitope prediction process into distinct components: (1) antigen processing through specific pathways (autophagy, phagocytosis, proteasomal degradation), (2) HLA class II molecule loading in endosomal compartments, (3) allele-specific binding preferences, and (4) peptide-MHC stability. Each segment is modeled separately with appropriate processing rules, then integrated to provide comprehensive and accurate predictions
Data Source
AI summary
Methods for preparing a personalized cancer vaccine and a method to train a machine learning HLA-peptide presentation prediction model. Further wherein, a method of making a HLA class II tetramer or multimer comprising an epitope, the method comprising contacting a purified soluble HLA-DM loaded with a peptide epitope with a HLA class II tetramer or multimer, thereby forming a HLA class II tetramer or multimer loaded with the peptide epitope, is disclosed.