Predictive Feature Set Selection Using Genetic Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to determine reliable sets of features that predict phenotypic characteristics, particularly in medical contexts, due to the challenge of systematically testing physiological and genetic features in real-life patients.
Innovation Solution
A computer-implemented method using a genetic algorithm to generate large populations of feature profiles, cluster them, and select subsets based on fitness scores to identify predictive feature sets, followed by a clustering process to extract useful predictors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine-learning algorithms are trained to predict phenotypic characteristics, then predictive accuracy improves, but the complexity of determining which features to use increases
Solution Approach 1:
The patent creates virtual copies of patients (synthetic training data) that replicate the complex relationships between features and phenotypic characteristics. These virtual patients are generated by sampling from the distribution of real patients, allowing the system to train machine-learning algorithms on numerous similar cases without working with the actual complex real-world data directly, thus simplifying the feature selection process while maintaining predictive accuracy.
Solution Approach 2:
The patent introduces virtual patients as an intermediary between real patient data and the machine-learning model. These virtual patients serve as a bridge that simplifies the training process by providing a manageable representation of patient data that captures essential relationships without the full complexity of real-world datasets, enabling easier feature identification and selection.
2Adaptability or versatility
If the number of physiological and genetic features to be tested increases, then the comprehensiveness of prediction improves, but the difficulty of systematically testing these features increases
Solution Approach 1:
The patent generates virtual patients that copy the feature distributions and relationships of real patients. By creating numerous virtual patients with varied feature profiles, the system can systematically test many physiological and genetic features without having to manually analyze each feature across the entire real patient population, thus reducing the difficulty of systematic testing while maintaining comprehensive prediction capability.
Solution Approach 2:
The patent performs preliminary analysis by generating virtual patients before conducting systematic feature testing. This preliminary step creates a simplified training dataset that pre-organizes the complexity of real patient data, allowing subsequent systematic testing of features to be performed more easily on the pre-processed virtual data rather than directly on the complex real-world data.
3Reliability
If real-life patient data is used for feature testing, then the relevance of predictions improves, but the ability to control and systematically test features is lost
Solution Approach 1:
The patent creates virtual patients that copy the essential characteristics and feature distributions of real patients. These virtual patients maintain the relevance of predictions by reflecting real patient patterns, while simultaneously providing the controllability needed for systematic feature testing through their generated nature, allowing researchers to deliberately vary and control feature combinations without the constraints of actual patient data.
Solution Approach 2:
Instead of starting with real patient data and attempting to control which features are tested (which is impossible), the patent inverts the approach by generating virtual patients first with controlled feature sets, then using these virtual patients to train models that can subsequently predict real patient outcomes. This inversion allows systematic testing of features while maintaining prediction relevance.
Data Source
AI summary
A computer-implemented method of determining one or more sets of features to predict the presence of a particular phenotypic characteristic comprises: (a) receiving patient data comprising, for each of a plurality of patients: a feature profile comprising a respective feature status for each of a plurality of features for that patient; and an indication of whether that patient expresses the particular phenotypic characteristic; (b) using a genetic algorithm to generate a plurality of generations of individuals, wherein each individual comprises a subset of the predetermined plurality of features, each generation of individuals generated based, at least in part, on a plurality of fitness scores, each fitness score corresponding to a respective individual in the previous generation, and parameterizing a predictive accuracy of the set of features, each fitness score being calculated based at least in part on the patient data; (c) repeating step (b) until it has been performed N times; (d) from the plurality of individuals generated in steps (b) and (c), selecting a subset of the individuals based on their fitness scores; (e) clustering the selected subset of individuals to generate a plurality of clusters of individuals, based on the similarity of their respective subsets of features; (f) from each cluster, identifying a respective characteristic feature set based on the frequency with which features appear in individuals in that cluster.