RelSeg

Relevance Based Segmentation of Nanopore Reads

RelSeg is used to align the basecalled sequence to the signal of nanopore reads. It relies on the bonito basecaller of ONT. The lxt and zennit packages are used for the Layer-wise Relevance Propagation.
A transformer model which no longer requires flash_attn is implemented.

Installation

$ pip install relseg

Usage

$ relseg rna004_130bps_sup@v5.0.0 /path/data/reads --rna > basecall.txt

$ relseg rna004_130bps_sup@v5.0.0 /path/data/reads --rna --save_relevance > basecall.txt

Output

Example Output

read_id	base	start	end
5a729d16-b785-4e8c-ad91-314d862d980b	T	140	157
5a729d16-b785-4e8c-ad91-314d862d980b	C	157	157
5a729d16-b785-4e8c-ad91-314d862d980b	T	157	181
5a729d16-b785-4e8c-ad91-314d862d980b	C	-1	-1

The normal basecalls including the moves table are also output.

Relevance for Consecutive Bases

Sequence Aligned to Signal