Sequence Variant Detection with Deep Learning Embeddings for Low-Coverage Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting genetic variations in sequencing data, particularly from suboptimal datasets, suffer from high error rates and require labor-intensive processing, limiting their accuracy and efficiency.
Innovation Solution
The use of embeddings and deep learning models to process biopolymer sequence reads, allowing for the detection of candidate variations, even in low-coverage data, by generating embeddings of sequence reads and applying convolutional neural networks for accurate classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If classical bioinformatic methods and machine learning modeling are used for variant calling, then processing can be performed with existing tools, but labor-intensive hand-crafting of data features is required and accuracy deteriorates on suboptimal datasets
Solution Approach 1:
The patent replaces manual feature engineering (mechanical process) with deep learning models that automatically learn optimal features from raw sequencing data. The neural network architecture (CNNs, LSTMs, Transformers) substitutes for hand-crafted feature extraction, eliminating labor-intensive processes while improving accuracy on suboptimal datasets through end-to-end learning.
Solution Approach 2:
The patent changes the approach from fixed, hand-selected features to dynamically learned parameters through deep learning. The model automatically adjusts feature representations based on data characteristics, allowing adaptation to different sequencing depths and qualities without manual reconfiguration, thus maintaining high accuracy across varying data conditions.
2Measurement precision
If deep learning models are used to detect variations in sequencing data, then accuracy improves especially for suboptimal datasets, but processing time and computational resources increase
Solution Approach 1:
The patent segments the sequencing data into manageable units (reads, variants, genomic regions) and processes them through modular deep learning components. This allows parallel processing and selective computation based on data characteristics, reducing overall processing time while maintaining accuracy through focused analysis of critical regions.
Solution Approach 2:
The patent implements dynamic processing where the deep learning model adjusts its computational intensity based on data quality and complexity. The system can operate at different resolution levels and adaptively allocate resources, speeding up processing for high-quality data while maintaining thorough analysis for suboptimal datasets without unnecessary computational overhead.
3Productivity
If high-throughput sequencing is used to sequence genomes rapidly, then cost and time are reduced, but error rates increase to 0.1-10%
Solution Approach 1:
The patent incorporates feedback mechanisms where the deep learning model continuously refines its predictions by analyzing patterns across multiple reads and comparing them against the reference genome. The model learns from discrepancies between expected and observed sequences, automatically adjusting its error correction to maintain high reliability even when raw sequencing error rates are elevated.
Solution Approach 2:
The patent performs preliminary error correction and quality assessment using deep learning models before final variant calling. The system pre-processes the data to identify and correct obvious errors, filter low-quality reads, and prepare cleaned datasets for analysis, thereby improving reliability early in the pipeline without sacrificing the rapid throughput enabled by high-throughput sequencing.
Data Source
AI summary
Methods, systems, and computer program products are disclosed that use embeddings of candidate variation information and deep learning models to accurately and efficiently detect variations in biopolymer sequencing data, particularly suboptimal sequencing data.