Generative Adversarial Networks for Functional Protein Sequence Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for generating functional protein sequences are inefficient due to the vastness of protein sequence space, where only a tiny fraction folds correctly, and existing deep learning approaches fail to ensure correct folding and chemical activity of generated proteins.
Innovation Solution
A method using generative adversarial networks to select and process existing protein sequences, approximate the distribution of amino acids, and generate protein sequences that are likely to be functional, incorporating steps like selection, processing, and post-processing to ensure high functionality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If random mutagenesis is used to navigate protein sequence space, then protein variants can be generated, but protein fitness declines exponentially with each random mutation making functional protein discovery extremely inefficient
Solution Approach 1:
The patent replaces random experimental mutagenesis with a computational machine learning system that predicts functional protein sequences. The machine learning model uses neural networks to navigate sequence space and generate proteins with desired functions, substituting the mechanical/random process with an intelligent prediction system that maintains fitness while exploring variants.
Solution Approach 2:
The patent creates synthetic protein sequences that copy and combine features from natural functional proteins. By learning from existing functional sequences and generating new variants that preserve essential functional characteristics, the system efficiently explores sequence space without the fitness decline associated with random mutations.
2Reliability
If experimental screening techniques are used to test protein variants, then functional proteins can be identified, but the method is limited to testing only 10^6-9 variants due to the vast sequence space
Solution Approach 1:
The patent replaces physical experimental screening with computational prediction using machine learning models. The neural network system can evaluate and predict functionality of protein sequences in silico, enabling testing of vastly more variants than physical methods allow without requiring actual laboratory screening of each variant.
Solution Approach 2:
The machine learning model acts as an intermediary between sequence generation and functional testing. Instead of directly testing physical protein variants, the system uses computational predictions to identify functional sequences, serving as a mediator that expands the searchable space beyond experimental capabilities.
3Productivity
If recombination of naturally occurring homologous proteins is used, then functional proteins with many mutations can be generated in a single step, but the method is strongly limited by the number of available parent molecules
Solution Approach 1:
The patent uses machine learning to copy and recombine features from natural protein sequences to generate synthetic variants. The neural network learns from existing functional proteins and creates new sequences that combine beneficial features, enabling exploration of sequence space beyond what is limited by available natural parents.
Solution Approach 2:
The patent transitions from exploring sequence space through recombination of existing natural proteins to generating sequences in a higher dimensional space of possibilities. The machine learning model can create variants that combine features from multiple natural proteins in ways not possible through traditional recombination, effectively adding a computational dimension to sequence exploration.
4Productivity
If deep generative algorithms using autoregressive neural networks are used, then protein sequences can be generated, but the methods do not ensure the correct folding or chemical activity of the generated proteins
Solution Approach 1:
The patent incorporates feedback mechanisms where the machine learning model continuously refines its predictions based on evaluation of generated sequences. The system uses feedback from functional assays and structural analysis to improve the accuracy of generated proteins, ensuring correct folding and chemical activity through iterative refinement rather than single-pass generation.
Data Source
AI summary
The invention generally relates to the field of protein sequences and of generation of functional protein sequences. More particularly, the invention concerns a method for generating functional protein sequences with generative adversarial networks. The described method for functional sequence generation comprises plurality of steps, each of which is crucial to ensure the high percentage of functional sequences in the final sequence set: selecting a plurality of existing protein sequences to define the approximate sequence space for the later generated synthetic sequences, processing the selected protein sequences, approximating the unknown true distribution of amino acids of the pre-processed sequences using a variation of generative adversarial networks, obtaining protein sequences from the approximated distribution, processing of the obtained protein sequences. The described method provides a resource (e.g. time, cost) efficient way of producing synthetic protein sequences which have a high probability of being functional experimentally.