MiniScrub

MiniScrub removes low-quality segments from Oxford Nanopore long reads using a convolutional neural network to improve read accuracy and reduce mis-assemblies for de novo genome assembly and structural variation analysis.

Key Features:

• Convolutional Neural Network (CNN): Employs a CNN to predict and enable removal of low-quality segments within Nanopore reads.
• Read-to-Read Overlap Generation: Uses MiniMap2 to generate read-to-read overlaps that inform segment-level quality assessment.
• Image Encoding: Encodes read-to-read overlaps into image representations for CNN input.
• Robust Performance: Demonstrates consistent improvement in read quality across control datasets and parameter settings.
• Indel Error Reduction: Produces fewer large indel errors relative to raw Nanopore reads.

Scientific Applications:

• De novo Genome Assembly: Improves the accuracy of de novo assemblies by removing low-quality read segments that can cause mis-assemblies.
• Metagenomic Analyses: Enhances read quality in metagenomic contexts to aid assembly and downstream analyses in mixed-species samples.
• Structural Variation Identification: Reduces large indel errors and mis-assemblies to facilitate more reliable structural variation detection.

Methodology:

Generate read-to-read overlaps with MiniMap2; encode overlaps into image formats; process images with a convolutional neural network to predict low-quality segments; remove predicted low-quality segments from reads.