Antibody Sequence Generation Using Similarity-Guided Amino Acid Mutation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods struggle to generate diverse antibody sequences while maintaining high binding affinity, often resulting in low meeting rates in actual experiments due to inappropriate or drastic changes in amino acid configurations.
Innovation Solution
A sequence generation device and method that includes a sequence proxy evaluation unit and a sequence generation unit, which uses a learning model to change amino acids with a predetermined probability, selecting amino acids with a higher probability for characteristics similar to the original, and trains the model to enhance proxy evaluation values, thereby generating sequences with improved binding affinity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If amino acids are randomly changed to increase sequence diversity, then sequence variety is improved, but binding affinity deteriorates
Solution Approach 1:
The patent applies local quality by differentiating the treatment of amino acids based on their position and functional importance. Conservative mutations are applied to less critical positions while preserving critical residues, allowing sequence diversity without compromising binding affinity. The probability density function assigns different mutation probabilities to different amino acid positions based on their functional significance.
Solution Approach 2:
The patent changes the parameter of mutation probability from uniform random selection to a probability density function that varies based on amino acid characteristics and position. This allows controlled diversity by adjusting the shape and parameters of the probability distribution to balance sequence variety with affinity preservation.
2Measurement precision
If sequences are generated with high binding affinity through simulations, then simulation evaluation is improved, but actual experiment performance deteriorates due to unaccounted evaluation items
Solution Approach 1:
The patent performs preliminary actions by generating and evaluating sequences through simulations before actual experiments. The probability density function is designed to pre-account for multiple evaluation items (binding affinity, viscosity, solubility, immunogenicity) by incorporating their constraints into the generation process, so that sequences meeting simulation criteria are more likely to succeed in actual experiments.
Solution Approach 2:
The patent implements feedback by using the results of simulation evaluations to refine the probability density function and generation parameters. The feedback loop ensures that sequences generated based on simulation performance are iteratively improved to better predict actual experimental outcomes, addressing the gap between simulated and actual performance.
3Reliability
If reinforcement learning is used to generate sequences with high binding affinity, then evaluation value is improved, but sequence diversity deteriorates due to convergence to similar configurations
Solution Approach 1:
The patent changes the parameter of selection from monotonous Q-function guidance to a probability density function that allows exploration of multiple modes. This enables the system to generate sequences with high binding affinity while maintaining diversity by sampling from a distribution rather than converging to a single optimal path.
Solution Approach 2:
The patent introduces dynamics by using a probability density function that can adaptively explore different sequence spaces. The dynamic nature of the probability distribution allows the system to transition between exploration of diverse sequences and exploitation of high-affinity configurations, preventing premature convergence while maintaining affinity.
Data Source
Figure 1~2
Figure 3
Figure 4~5(b)
AI summary
A sequence generation device according to the present disclosure includes a sequence proxy evaluation unit that calculates a proxy evaluation value for a sequence including a plurality of amino acids, and a sequence generation unit that generates the sequence using a learning model, in order to generate various sequences while maintaining high binding affinity. The sequence generation unit changes some of the amino acids in the sequence generated by the learning model with a predetermined probability, selects an amino acid after change using a probability density function such that the amino acid having a characteristic closer to that of the amino acid before change is selected with a higher probability, trains the learning model to generate a sequence indicating a higher proxy evaluation value based on the proxy evaluation value for the sequence after the amino acid is changed, thereby generating a trained model, and generates a generation sequence using the trained model (see FIG. 3).