Iterative Clustering for Long Read Error Correction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current genome aligners are inadequate for error correction of full-length long sequence reads in transcriptome sequencing, particularly when sequencing technologies like PacBio generate reads with high error profiles, and existing methods relying on short reads introduce additional errors and biases, failing to utilize the redundancy and quality values of long reads.

Innovation Solution

An iterative clustering method for sequence reads is employed, where reads are grouped based on similarity, cluster consensuses are generated, and the clustering is iteratively improved using quality values to reassign reads and merge similar clusters, enabling accurate error correction without a reference genome.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If short reads are used to correct long reads through alignment, then error correction is attempted, but short reads map nonspecifically and introduce more errors

Engineering Contradiction:
Improveerror correction accuracyVSAvoidnonspecific mapping errors
Core Design Contradiction:
Measurement precisionVSObject-generated harmful factors

Solution Approach 1:

Instead of using short reads to correct long reads, the patent inverts the approach by using long reads to correct short reads. The long reads serve as the reference scaffold, and short reads are aligned against them for correction, reversing the traditional error correction paradigm to eliminate nonspecific mapping issues.

Inventive Principle:
Principle #13The other way round (Inversion)

Solution Approach 2:

The patent creates a consensus sequence copy from multiple long reads that serves as a high-accuracy reference. This consensus copy is then used to correct individual long reads and to guide short read alignment, eliminating the need to map short reads directly against each other or against a potentially inaccurate reference.

Inventive Principle:
Principle #26Copying

2Measurement precision

If existing error correction tools are used, then some error correction is achieved, but they require two different sequencing systems and introduce systematic biases

Engineering Contradiction:
Improveerror correction accuracyVSAvoidsequencing system compatibility
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal error correction framework that works with any sequencing system that produces long reads. The method uses the long reads themselves as the correction source, eliminating dependence on specific short read technologies and their associated systematic biases, making the approach system-agnostic.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent converts the previously harmful high error rate of long reads into a benefit by using multiple long reads of the same transcript to generate a consensus sequence. The redundancy in long read sequencing is transformed into a strength, where the very reads that contained errors now provide the data needed to create a highly accurate reference.

Inventive Principle:
Principle #22Blessing in disguise (Convert harm into benefit)

3Loss of information

If long reads are used for transcriptome sequencing, then full-length transcripts can be sequenced to identify splicing patterns, but the reads contain high error profiles that prevent direct alignment

Engineering Contradiction:
Improvetranscriptome information completenessVSAvoidsequence accuracy
Core Design Contradiction:
Loss of informationVSMeasurement precision

Solution Approach 1:

The patent performs preliminary error correction by generating a consensus sequence from multiple long reads before attempting alignment or analysis. This pre-correction step ensures that the long reads are sufficiently accurate for downstream applications while preserving the full-length information that makes long read sequencing valuable.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent merges multiple long reads of the same transcript to create a consensus sequence. By combining information from multiple reads, the method achieves both high accuracy (through error averaging) and full-length coverage (through the inherent length of long reads), resolving the contradiction between information completeness and sequence accuracy.

Inventive Principle:
Principle #5Merging (Combining)

4Productivity

If simple string matching algorithms are used, then accurate reads can be compared efficiently, but real-world sequencing data contains errors that make simple matching insufficient

Engineering Contradiction:
Improvesequence comparison efficiencyVSAvoidvariant identification accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent introduces a consensus sequence as an intermediary between the raw long reads and the variant identification process. This mediator serves as a high-accuracy reference that enables efficient comparison while maintaining precision, as the consensus has already had errors corrected through the merging of multiple reads.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP3084426B1Iterative clustering of sequence reads for error correction
Publication Date: 2020.04.15 PACIFIC BIOSCIENCES OF CALIFORNIA INC
  • EP3084426B1 patent drawingFigure 1
  • EP3084426B1 patent drawingFigure 2
  • EP3084426B1 patent drawingFigure 3

AI summary

Exemplary embodiments provide methods and systems for iterative clustering of sequence reads for error correction. Aspects of the exemplary embodiment include receiving a set of sequence reads and associated quality values; grouping the sequence reads into a set of initial clusters based on sequence similarity; generating a cluster consensus for each of the initial clusters; iteratively improving the clustering based on the cluster consensus and the quality values associated with the sequence reads; and generating and outputting a final cluster consensus for each of the clusters.