Compound Coverage Scoring for Active Learning in Drug Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing drug discovery process is lengthy, expensive, and inefficient due to the synthesis and testing of numerous compounds, with computational methods struggling to accurately select compounds for training sets that effectively optimize ML models for predicting desired properties, as chemical space is not equally spaced and prior art metrics are less effective.
Innovation Solution
A computational method and system for selecting compounds in a chemical space to optimize training of a ML model by determining a subset score based on molecular properties' frequencies in the population and sampled set, using a coverage score that balances Shannon entropy and feature overlap, enabling improved exploration and exploitation strategies for compound selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional drug discovery methods are used to synthesize and test numerous compounds, then the likelihood of finding candidate compounds is improved, but the time and cost increase significantly
Solution Approach 1:
The patent applies preliminary action by using computational methods to pre-evaluate and rank compounds before synthesis. A machine learning model predicts biological activity and a coverage score metric evaluates diversity of molecular properties, allowing researchers to select the most promising compounds for synthesis in advance, thereby reducing the number of design cycles needed.
Solution Approach 2:
The patent replaces the mechanical system of synthesizing and testing compounds in the laboratory with a computational system that uses machine learning models and coverage score calculations. This substitution allows for rapid evaluation of compound properties without physical synthesis, significantly reducing time and cost while maintaining reliability in identifying candidate compounds.
2Loss of information
If more compounds are synthesized and tested, then the training set for ML models becomes more comprehensive, but the cost and time of the project increase
Solution Approach 1:
The patent applies self-service by using the coverage score metric to automatically evaluate and select compounds that provide the most valuable information for training. The system self-determines which compounds to synthesize based on their molecular property diversity and predicted activity, eliminating the need to synthesize all possible compounds while still achieving comprehensive training data.
Solution Approach 2:
The patent changes the parameter of compound selection from random or activity-only based to a dual-criteria approach using coverage score (molecular property diversity) and predicted biological activity. This parameter change allows the system to select a smaller number of compounds that maximize information content for training, reducing both cost and time while maintaining training set comprehensiveness.
3Difficulty of detecting and measuring
If computational methods are used to predict compound properties, then the analysis depth increases, but the accuracy of predictions depends on the quality of training data
Solution Approach 1:
The patent applies feedback by using the coverage score metric to evaluate the diversity of molecular properties in the training set and adjust compound selection accordingly. This feedback mechanism ensures that the training data is both comprehensive and representative, which improves the accuracy of machine learning predictions while maintaining deep analysis capabilities.
Solution Approach 2:
The patent applies preliminary action by pre-evaluating compound diversity and predicted activity before synthesis. The coverage score metric assesses molecular property distribution in advance, ensuring that the training set will be representative and high-quality, which directly improves prediction accuracy while enabling deep computational analysis.
Data Source
AI summary
A method for computational drug design includes defining a population of a plurality of compounds. Each compound includes one or more molecular properties. The method includes defining a training set of compounds from the population for which one or more biological properties are known. The method includes selecting, from the population, a subset of one or more compounds that are not in the training set. The method includes determining a subset score of the selected subset based on molecular properties of the one or more compounds in the selected subset, and evaluating the selected subset based on the determined subset score. The subset score is determined based on a frequency of the molecular properties in the population and on a frequency of the molecular properties in a sampled set comprising the training set and the selected subset.