RNN Attention Encoder for Variable-Length MHC-Peptide Binding Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current neural network models for predicting MHC-peptide binding affinity struggle with variable length peptide sequences, as they often require padding, which limits predictive performance and accuracy.

Innovation Solution

A neural network model employing a recurrent neural network encoder with attention weighting is used to process variable length peptide sequences, generating attention weighted outputs to improve prediction accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If neural network models use fixed-length inputs with padding for variable length peptide sequences, then the model structure remains simple and consistent, but the predictive performance and accuracy are limited

Engineering Contradiction:
Improvepredictive performanceVSAvoidmodel structure complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies dynamics by transitioning from fixed-length input processing to variable-length processing through recurrent neural networks. The RNN encoder dynamically adapts to different peptide sequence lengths without requiring padding, allowing the model to process sequences of varying lengths while maintaining optimal predictive performance. This dynamic approach resolves the contradiction by enabling length-flexible processing without sacrificing model consistency.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the fundamental parameter of input length handling from fixed to variable. By implementing an RNN-based encoder that naturally accommodates variable-length sequences, the model transforms the input processing mechanism to eliminate the need for padding operations. This parameter change enables the system to achieve higher predictive performance while maintaining architectural simplicity through the inherent sequential processing capability of RNNs.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If padding is applied to match fixed input length, then the neural network can process variable length sequences, but the full predictive performance potential cannot be achieved

Engineering Contradiction:
Improvevariable length processing capabilityVSAvoidpredictive accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The RNN encoder provides dynamic length adaptation by processing sequences step-by-step, automatically adjusting to any input length without padding. This dynamic processing mechanism maintains both adaptability to variable lengths and full predictive performance potential, as the model can process each amino acid position sequentially and capture position-specific information effectively.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent introduces an attention mechanism as an intermediary between the RNN encoder and the prediction layer. This attention mechanism weights the importance of different peptide positions, allowing the model to focus on critical binding regions while maintaining variable-length processing capability. The attention weights serve as intermediaries that preserve predictive accuracy by highlighting position-specific contributions to MHC binding affinity.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Ease of manufacture

If traditional neural networks are used with padding, then implementation is straightforward, but the ability to determine important peptide positions is limited

Engineering Contradiction:
Improveimplementation simplicityVSAvoidpeptide position importance information
Core Design Contradiction:
Ease of manufactureVSLoss of information

Solution Approach 1:

The attention mechanism acts as an intermediary that extracts and highlights position-specific importance information from the peptide sequence. By computing attention weights for each amino acid position, the model identifies which positions contribute most to MHC binding affinity. This intermediary mechanism preserves complete position information without loss, while the attention weights provide interpretable insights into critical binding regions.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The attention mechanism can be visualized as providing different 'weights' or 'intensities' to different peptide positions, analogous to color changes highlighting important features. Positions with higher attention weights are effectively 'highlighted' as more important for binding, allowing researchers to identify critical peptide regions. This visualization approach maintains implementation feasibility while revealing position-specific importance information.

Inventive Principle:
Principle #32Color changes

Data Source

PatentEP4550338B1Methods and systems for improved major histocompatibility complex (MHC)-peptide binding prediction of neoepitopes using a recurrent neural network encoder and attention weighting
Publication Date: 2026.04.01 NANTOMICS LLC
  • EP4550338B1 patent drawingFigure 1
  • EP4550338B1 patent drawingFigure 2
  • EP4550338B1 patent drawingFigure 3

AI summary

Techniques are provided for predicting MHC-peptide binding affinity. A plurality of training peptide sequences is obtained, and a neural network model is trained to predict MHC-peptide binding affinity using the training peptide sequences. An encoder of the neural network model comprising an RNN is configured to process an input training peptide sequence to generate a fixed-dimension encoding output by applying a final hidden state of the RNN at intermediate state outputs of the RNN to generate attention weighted outputs, and linearly combining the attention weighted outputs. A fully connected layer following the encoder is configured to process the fixed-dimension encoding output to generate an MHC-peptide binding affinity prediction output. A computing device is configured to use the trained neural network to predict MHC-peptide binding affinity for a test peptide sequence.