Consensus Sequence Generation Using LSTM Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sequencing techniques, such as third-generation and next-generation sequencing, suffer from high error rates, often reaching up to 40%, which hampers the utility of sequencing in identifying diseases and characterizing therapies.
Innovation Solution
A computer-implemented method is provided to generate a consensus sequence by accessing a set of independently generated sequences, performing multiple sequence alignment, and using a machine learning model, specifically a recurrent neural network with long short-term memory (LSTM) units, to process feature vectors and reduce errors by collapsing homopolymers and incorporating quality metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If existing sequencing techniques (third-generation or next-generation) are used, then sequencing can be performed rapidly and in real-time, but the error rate reaches up to 40%
Solution Approach 1:
The patent combines multiple independently generated sequences into a single consensus sequence by aligning them and selecting the most reliable base calls. This merging approach allows the system to maintain the rapid sequencing speed of third-generation techniques while reducing the error rate from 40% to below 1% through collective validation across multiple sequences.
Solution Approach 2:
The alignment process provides feedback by comparing each base position across multiple sequences and using quality metrics to identify and correct errors. The system uses this feedback loop to generate a consensus sequence that reflects the most accurate reconstruction of the original molecule, effectively reducing errors while preserving sequencing speed.
2Measurement precision
If multiple sequences are aligned and processed using machine learning models, then sequencing accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the computational task by processing sequences in an alignment framework where each position is evaluated independently using feature vectors. This segmentation allows the machine learning model to focus on local patterns at each base position rather than analyzing entire sequences globally, reducing computational complexity while maintaining high accuracy through position-specific feature extraction.
Solution Approach 2:
The patent introduces feature vectors as an intermediary representation between the raw multiple sequences and the machine learning model. These feature vectors capture essential information from each sequence at each position, transforming complex sequence data into a simplified format that the LSTM model can process efficiently, thereby reducing computational complexity while preserving sequencing accuracy.
3Measurement precision
If homopolymers are collapsed to single bases, then alignment accuracy is improved, but information about homopolymer length is lost
Solution Approach 1:
The patent uses feature vectors as an intermediary that preserves homopolymer length information while enabling accurate alignment. Instead of collapsing homopolymers to single bases, the feature vectors encode the length and composition of homopolymer regions, allowing the alignment process to maintain accuracy while the machine learning model recovers the original homopolymer structure in the consensus sequence.
Solution Approach 2:
The patent changes the representation parameter from collapsed single bases to feature vectors that include homopolymer length information. This parameter change allows the system to maintain alignment accuracy by providing explicit length information to the machine learning model, which then uses this information to correctly reconstruct homopolymer regions in the consensus sequence without losing critical length data.
Data Source
AI summary
Some embodiments relate to methods, systems, uses, or software for generating a consensus sequence of a particular molecule. A set of sequences of the particular molecule can be accessed, each having been generated independently from other sequences in the set of sequences and each including an ordered set of bases. An alignment process may be performed using the set of sequences to generate an alignment result associating, for each base of the ordered sets of bases of the sets of sequences. The base may have a reference position. For each reference position of a set of reference positions, a feature vector for the reference position may be generated that represents each base from the ordered sets of bases aligned to the reference position. The feature vectors for the set of references positions may be processed using a machine learning model to generate the consensus sequence for the particular molecule.