Deep Learning Variant Calling for Accurate Gene Carrier Status
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for identifying genetic conditions associated with the number of functional copies of specific genes and single nucleotide polymorphisms require significant human involvement and lack accuracy in determining carrier statuses.
Innovation Solution
Utilizing machine learning algorithms, particularly deep learning and recurrent neural networks (RNNs), to analyze copy number and SNP data from genomic sequencing to accurately determine the number of functional copies of genes and identify genetic conditions, incorporating anomaly detection to ensure data quality and confidence thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used to analyze genomic data, then measurement precision of carrier status determination is improved, but device complexity increases
Solution Approach 1:
The machine learning system is divided into distinct modular components: an anomaly detection module that preprocesses genomic data to identify and flag abnormal patterns, and a recurrent neural network module that processes the data sequentially to determine carrier status. This segmentation allows each module to specialize in specific tasks, improving overall measurement precision while making the complex system more manageable and interpretable.
Solution Approach 2:
The patent introduces confidence scores as an intermediary metric between the neural network's raw output and the final carrier status determination. These confidence scores provide an additional layer of validation and interpretation, allowing the system to handle uncertainty appropriately and improving the reliability of the final diagnostic output without requiring complete redesign of the underlying complex algorithms.
2Productivity
If deep learning models are implemented for variant calling, then productivity of genetic condition identification is improved, but ease of operation deteriorates
Solution Approach 1:
The recurrent neural network is designed to automatically process sequential genomic data and generate carrier status determinations without requiring manual intervention or complex configuration. The system self-adjusts by learning from training data and automatically handles the complex task of analyzing copy number and SNP data, thereby improving productivity while maintaining ease of operation through automation.
Solution Approach 2:
The system performs preliminary anomaly detection and data preprocessing before the main carrier status determination step. By pre-processing the genomic data to identify and handle anomalies upfront, the system streamlines the subsequent analysis process, improving overall productivity while making the main operation simpler and more reliable.
3Reliability
If anomaly detection is incorporated into the sequencing analysis, then reliability of results is improved, but device complexity increases
Solution Approach 1:
The anomaly detection functionality is merged directly into the recurrent neural network processing pipeline rather than being implemented as a separate external system. This integration allows the anomaly detection to work seamlessly with the main analysis workflow, improving reliability through comprehensive quality control while minimizing the increase in overall system complexity through unified architecture.
Data Source
AI summary
A system and method for determining a respective carrier status of an individual is provided. In one implementation, a method includes training a neural network model based on predetermined information related to at least one genetic variant and determining the respective carrier status based on a normalized read depth for the gene in a genome of the individual and allele dosage data for the gene using a machine learning algorithm. The method is configured to receive, as inputs, the normalized read depth and allele dosage data, and output the respective carrier status of the individual for the at least one genetic variant. In another implementation a system includes a recurrent neural network. The recurrent neural network includes a plurality of long short-term memory (LSTM) network cells linked sequentially in the recurrent neural network, a plurality of layers of long short-term memory (LSTM) network cells; and a fully connected sigmoid module.