Dual-Branch Neural Network for Enzymatic Reaction Feasibility Evaluation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for evaluating enzymatic reaction feasibility in bio-retrosynthesis pathway planning are inefficient, as they rely heavily on human expertise and do not effectively utilize rich sequence features in molecular SMILES character strings, leading to high experimental costs and inaccurate feasibility evaluations.
Innovation Solution
A dual-branch feature extraction network based on a convolutional neural network and attention mechanism is established, incorporating multi-modal features from substrate and product molecules, with a product SMILES sequence generation task to enhance feasibility classification, allowing for more accurate evaluation of enzymatic reaction feasibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a field expert determines enzymatic reaction feasibility by considering biochemical knowledge (energy changes, entropy changes, chemical bond breaking or formation, enzyme activity, chassis cell environment), then the determination is highly authoritative, but the process needs lots of professional knowledge and labor costs
Solution Approach 1:
The patent creates a computational model that copies and simulates the expert evaluation process. The model learns from training data containing enzymatic reaction examples and their feasibility outcomes, enabling it to automatically evaluate new reactions without requiring human expert intervention for each case, thus reducing labor costs while maintaining reliable assessments
Solution Approach 2:
The patent replaces the manual mechanical process of expert evaluation with an automated computational system. The model uses machine learning algorithms to process molecular structures and predict reaction feasibility, substituting the human expert's cognitive process with an automated computational mechanism that operates faster and at lower cost
2Extent of automation
If existing machine learning methods evaluate enzymatic reaction feasibility, then the process is automated, but the methods do not consider rich sequence features included in molecular SMILES character strings and merely regard model training as a binary classification task, resulting in insufficient accuracy and reliability
Solution Approach 1:
The patent transitions from treating SMILES strings as simple sequences to processing them as multi-dimensional features. The model incorporates both the sequential character information and the structural molecular information represented by SMILES, creating a more comprehensive feature space that captures richer patterns for improved accuracy
Solution Approach 2:
The patent changes the modeling approach from simple binary classification to a more sophisticated multi-task learning framework. The model simultaneously performs sequence generation and feasibility classification, using multiple loss functions and optimization objectives that better capture the complexity of enzymatic reaction evaluation, thereby improving both accuracy and reliability
3Adaptability or versatility
If a large number of enzymatic reactions are derived from bio-retrosynthesis pathway planning, then comprehensive pathway options are available, but explosion of various possible combinations occurs, making it difficult for even experienced biologists to select the most likely reaction
Solution Approach 1:
The patent enables the system to automatically evaluate and rank enzymatic reactions without human intervention. The trained model independently processes each reaction in the derived set, assigns feasibility scores, and identifies the most promising candidates, making the system self-sufficient in handling the combinatorial explosion problem
Solution Approach 2:
The patent introduces a computational evaluation model as an intermediary between pathway planning and reaction selection. This intermediary automatically filters and ranks the large number of derived reactions based on learned patterns from training data, bridging the gap between comprehensive pathway generation and practical reaction selection by providing objective feasibility assessments
Data Source
AI summary
Provided is a method and a system for evaluating enzymatic reaction feasibility based on multiple tasks and molecular multi-modal features. An enzymatic reaction feasibility dataset is constructed with a public dataset and a bioengineering reaction rule template library; SMILES sequence features and Morgan fingerprint spatial structure features of a product molecule and a substrate molecule of a reaction are used as inputs to a neural network; a dual-branch network is constructed based on an attention mechanism and a convolutional neural network to extract molecular multi-modal features; a product SMILES sequence generation task is taken as a secondary task to strengthen the capability of a model learning sequence feature; richer features are provided for an enzymatic reaction feasibility evaluation task; and the trained model is effectively enabled to accurately determine reaction feasibility by taking the molecular multi-modal features into overall consideration.