Nubeam

Nubeam encodes short sequencing reads as numerical values via matrix-based representations to enable reference-free comparison of metagenomic and other sequencing samples.

Key Features:

• Reference-free analysis: Compares sequencing samples without relying on reference genomes to avoid reference and mapping bias.
• Matrix representation of nucleotides: Represents each nucleotide using matrices and transforms reads into products of these matrices.
• Noncommutative differentiation: Exploits the noncommutative property of matrix multiplication so that similar reads receive similar numerical values while dissimilar reads are distinguished.
• Empirical distributions: Converts collections of read-assigned numbers into empirical distributions for each sample.
• Distance-based comparison: Quantifies genetic differences between samples by measuring distances between their empirical distributions.
• k-mer extension: Includes the k-mer method as a special case of its matrix-based representation.
• GC-bias and quality adjustment: Incorporates handling of GC bias and nucleotide quality in the analysis.
• WGS and unmapped reads: Applies to whole genome shotgun (WGS) sequencing data and explicitly accounts for unmapped reads.
• 16S rRNA applicability: Can be applied to 16S rRNA sequencing data for microbiota studies.
• Reproduction and insight: Recapitulates findings from mapping-based methods while revealing contributions from unmapped reads.

Scientific Applications:

• Metagenomic sample comparison: Enables reference-free comparison of metagenomic sequencing datasets to assess community differences.
• WGS analysis including unmapped reads: Analyzes whole genome shotgun datasets to study genetic variation when many reads remain unmapped.
• 16S rRNA microbiota studies: Supports 16S rRNA analyses such as assessing mouse gut microbiota resilience under stress and comparing vaginal microbiota in polycystic ovary syndrome versus healthy controls.
• Quantification of genetic differences: Provides a metric for quantifying genetic differences between samples via distributional distances.

Methodology:

Represents nucleotides as matrices, transforms each read into the product of those matrices, assigns numerical values using noncommutative matrix multiplication, constructs empirical distributions from read values, and measures distances between distributions; accounts for GC bias and nucleotide quality and treats the k-mer method as a special case.