Generative Model for Protein Humanization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for humanizing proteins to reduce immunogenicity while preserving desired functions are inefficient, as they fail to effectively generate protein sequences with altered immunogenicity while maintaining enzymatic activity or target binding affinity.
Innovation Solution
A method using a generative model, comprising an encoder neural network and a decoder neural network, that evaluates and weights protein sequences to generate sequences with altered immunogenicity, retraining the model iteratively to optimize for reduced immunogenicity while preserving the desired function, such as enzymatic activity or target binding affinity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If traditional humanization methods are used to reduce immunogenicity, then immunogenicity is reduced, but desired function (enzymatic activity or target binding affinity) is lost
Solution Approach 1:
The patent changes the parameters of the protein sequence by using a generative model to sample and weight sequences according to their predicted immunogenicity scores and functional scores, thereby finding sequences that optimize both immunogenicity reduction and functional preservation simultaneously
Solution Approach 2:
The patent implements feedback loops where the generative model is retrained iteratively using weighted sampling based on predicted immunogenicity and functional scores from oracle models, allowing the system to learn and improve sequence generation that balances both immunogenicity reduction and functional preservation
2Object-affected harmful factors
If protein sequences are heavily modified to reduce immunogenicity, then immunogenicity is reduced, but sequence similarity to functional templates decreases
Solution Approach 1:
The patent transforms the sequence modification approach from heavy random modification to targeted modification guided by the generative model, which samples sequences with optimized weighting based on both immunogenicity reduction and functional score preservation, thereby maintaining sequence similarity while reducing immunogenicity
3Manufacturing precision
If extensive sampling and evaluation of protein sequences is performed, then immunogenicity optimization is improved, but computational time and resources increase
Solution Approach 1:
The patent performs preliminary action by pre-training the generative model on a dataset of protein sequences and using oracle models to pre-evaluate functional scores before the main optimization process, thereby reducing the computational burden during iterative retraining and sampling phases
Solution Approach 2:
The patent uses copying by training the generative model to learn from a dataset of existing protein sequences, allowing it to generate new sequences that inherit functional characteristics from the training data without requiring exhaustive sampling and evaluation of all possible sequences
Data Source
AI summary
Humanizing proteins can be a laborious process, often involving trial and error or other non-systematic methods. To improve humanization, neural networks can be employed to generate new protein sequences having higher probabilities of being humanized. In an embodiment, a method includes evaluating the immunogenicity of a sampling of protein sequences. The method can include weighting the sampling of protein sequences from the generative model according to an estimated probability of a particular generated protein sequence having a deviation in immunogenicity than a particular percentile of immunogenicity of the sampling of protein sequences. The method can further include generating a protein sequence weighted sampling of protein sequences. The generated protein sequence representing a protein has an altered immunogenicity. Such a generated protein has a higher likelihood of being humanized.