Rapid Real-time Squiggle Classification for Read until using RawMap.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Real-time sequencing with Oxford Nanopore Technology (ONT) requires efficient filtering of target species.
• Existing Read Until pipelines struggle with speed and accuracy, especially for samples with high non-target read fractions (>99%).
• The baseline Guppy deep neural network is slow and classifies only ~60% of sequenced bases.

Purpose of the Study:

• To develop a faster and more accurate software filter for ONT's Read Until sequencing.
• To reduce sequencing time, cost, and computational overhead in microbial enrichment applications.
• To create a species-agnostic classifier for real-time DNA sequencing applications.

Main Methods:

• Developed RawMap, a CPU-only Read Until classifier operating in the squiggle space (raw electrical signal data).
• Employed a Support Vector Machine (SVM) trained on non-linear characteristics of ONT squiggle data to distinguish human from microbial DNA.
• Evaluated RawMap's performance against the baseline Guppy-based Read Until pipeline.

Main Results:

• RawMap achieved a 1327X speed improvement over the baseline pipeline.
• RawMap-augmented pipelines reduced sequencing time and cost by approximately 24%.
• Computing costs were reduced by 22%, and RawMap demonstrated species-agnostic classification capabilities.

Conclusions:

• RawMap offers a significant advancement in real-time sequencing efficiency for microbial applications.
• The classifier reduces costs and computational demands, making ONT sequencing more accessible.
• RawMap's species-agnostic nature and speed suggest potential applications beyond microbial identification, including viral load quantification.