Deep Learning DNA Mixture Classification via Sparse Matrices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for classifying individuals from DNA mixtures, particularly in forensic and biomedical research, face challenges due to imbalanced contributor ratios, misclassification issues, and the need for pre-selected features, limiting their accuracy and applicability across different disciplines.
Innovation Solution
A deep learning model using a 1-dimensional convolutional neural network with a sliding window trimming procedure processes next-generation sequencing data to generate sparse matrices, enabling robust and unbiased feature selection for accurate classification of DNA mixtures, achieving high accuracy across various disciplines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing classification methods are used for DNA mixtures, then the process is simpler, but the accuracy is insufficient and misclassification rate is high
Solution Approach 1:
The patent segments the DNA mixture classification problem into multiple processing stages: raw NGS data is divided into sequence reads, which are then processed through quality filtering, trimming, and alignment steps. The deep learning model itself is segmented into multiple convolutional layers, fully connected layers, and activation functions, allowing complex classification to be achieved through a series of simpler transformations that improve accuracy while managing complexity systematically
Solution Approach 2:
The patent transforms the classification problem from traditional feature-space analysis to a high-dimensional sequence space by using raw NGS reads as input. The deep learning model operates in this high-dimensional space, creating abstract representations that capture complex patterns invisible to traditional methods. This dimensional transformation enables the model to achieve 97% accuracy by leveraging patterns across multiple sequence positions and combinations simultaneously
2Measurement precision
If pre-selected features are used for training, then the model training is faster, but feature bias occurs and accuracy decreases
Solution Approach 1:
The deep learning model performs automatic feature selection and extraction without requiring manual domain knowledge or pre-curation. The convolutional layers automatically identify relevant sequence patterns and combinations during training, and the model adapts to different DNA mixture compositions autonomously. This self-service capability eliminates feature bias while maintaining training efficiency, as the model learns optimal features directly from the data rather than relying on pre-selected features that may be biased toward specific mixture types
Solution Approach 2:
The patent changes the fundamental parameters of the classification approach by using raw sequence reads with varying lengths and compositions as direct input, rather than transforming data into pre-selected features. The model processes sequences of different lengths through padding and trimming operations, allowing it to learn from the full distribution of sequence variations. This parameter change enables the system to adapt to diverse DNA mixture compositions without being constrained by fixed feature sets
3Measurement precision
If traditional ROC curves are used for evaluation, then the evaluation process is simpler, but the precision is not sufficient to evaluate model performance thoroughly
Solution Approach 1:
The patent replaces traditional ROC curve analysis with a direct accuracy metric based on predicted versus actual class labels. Instead of using the mechanical process of generating ROC curves and calculating AUC values, the evaluation simply compares predicted classifications against ground truth labels to compute accuracy directly. This substitution maintains evaluation thoroughness by focusing on the most relevant metric for classification performance while eliminating the time-consuming ROC generation process
Data Source
AI summary
A method for classifying individuals in mixtures of DNA is disclosed. The method comprises: Provide next-generation sequencing (NGS) data which comprises raw sequence reads originated from mixtures of DNA; performing a data processing procedure to generate a plurality of sparse matrix; and input the plurality of sparse matrix into a trained deep learning model installed on computers to classify individuals in the mixtures of DNA. In particular, the method is used to classify individuals in mixture of the DNAs from forensic dataset or whole exome sequencing dataset.