Genomic Variant Classification Using Machine Learning and ACMG Guidelines
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current predictive methods for classifying genomic variants as pathogenic or benign are inadequate, often resulting in uncertain classifications due to a lack of standardized approaches and low predictive quality, especially when they fail to meet minimum criteria set by guidelines like ACMG/AMP, leading to reduced reliability and accuracy in diagnostic interpretations.
Innovation Solution
A method that uses a machine-learning algorithm trained on genomic data and predefined pathogenicity/benignity criteria, including both statistical and patient-specific conditions, to classify variants by processing input information and determining an estimated probability of pathogenicity, with an optimized threshold to minimize false positives and improve precision and sensitivity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data-driven prediction tools are used to classify variants, then classification speed is improved, but classification accuracy deteriorates due to lack of standardized approaches
Solution Approach 1:
The patent combines data-driven machine learning algorithms with guideline-based criteria (ACMG/AMP) into a unified classification system. The machine learning model processes variant data while simultaneously evaluating against predefined pathogenicity criteria, merging the speed of computational methods with the accuracy of standardized guidelines to achieve both fast and accurate classification.
Solution Approach 2:
The system incorporates feedback mechanisms where the machine learning model's predictions are continuously validated against guideline criteria and clinical outcomes. This feedback loop allows the model to refine its classifications over time, improving accuracy while maintaining the speed advantages of automated computation.
2Productivity
If machine learning tools are used for variant classification, then processing efficiency is improved, but reliability deteriorates due to lack of standardized classification
Solution Approach 1:
The patent designs a universal classification system that can apply multiple evaluation dimensions simultaneously - machine learning predictions, guideline criteria assessment, and clinical validation. This multi-functional approach ensures that classification is both efficient (through automated processing) and reliable (through standardized validation against ACMG/AMP guidelines).
Solution Approach 2:
The system performs preliminary classification using machine learning algorithms, then immediately follows with validation against predefined guideline criteria before final classification. This preliminary action allows efficient initial sorting while ensuring reliability through subsequent standardized verification, preventing unreliable classifications from being accepted.
3Stability of the object's composition
If variant interpretation follows ACMG/AMP guidelines, then classification standardization is improved, but classification completeness deteriorates due to minimum criteria requirements
Solution Approach 1:
The patent adds an additional dimension to the ACMG/AMP guideline framework by incorporating machine learning-based probability assessments. Instead of relying solely on binary criterion meeting, the system evaluates variants across multiple dimensions - guideline criteria satisfaction, machine learning probability scores, and clinical context - enabling more complete classification of variants that meet minimum criteria but still require careful assessment.
Solution Approach 2:
The system dynamically adjusts the weight and interpretation of different criteria based on the specific variant characteristics and clinical context. This dynamic evaluation allows the system to maintain standardization through guideline-based approaches while improving completeness by adaptively assessing each variant's unique features beyond minimum requirements.
Data Source
AI summary
A method is for determining the pathogenicity/benignity of a genomic variant in connection with a given disease includes accessing genomic data in a list of the patient's genomic variants and for each variant detected, verifying whether or not the variant meets each predefined pathogenicity/benignity criteria. Each of such pathogenicity/benignity criterion is a proposition, which can be true or false, related to the variant for a previously known condition or a patient-specific condition. Input information is prepared for a trained algorithm using artificial intelligence and/or machine learning. The input information includes information related to the pathogenicity/benignity criteria associated with the level of evidence met by the variant. The input information is processed by the trained algorithm, to obtain an output information representative of the pathogenicity/benignity of each variant. The algorithm is trained in a preliminary step of training.