Neural network based basecallers output a move table that squigualiser can use to align the raw signals to reads. Squigualiser currently supports many basecallers including slow5-dorado, buttery-eel and ont-Guppy.
The first step is to run the basecaller to generate the required move table information along with the basecalled reads. An explanation of the move table information is available here
Run basecaller (slow5-dorado, buttery-eel or ont-Guppy)
# buttery-eel (tested with v0.2.2)
buttery-eel -g [GUPPY exe path] --config [DNA model] -i [INPUT] -o [OUTPUT] --port 5558 --use_tcp -x "cuda:all" --moves_out
e.g buttery-eel -g [GUPPY exe path] --config dna_r10.4.1_e8.2_400bps_sup.cfg -i input_reads.blow5 -o out.sam --port 5558 --use_tcp -x "cuda:all" --moves_out
# slow5-dorado (tested with v0.2.1)
slow5-dorado basecaller [DNA model] [INPUT] --emit-moves > [OUTPUT]
e.g. slow5-dorado basecaller dna_r10.4.1_e8.2_400bps_sup@v4.0.0 input_reads.blow5 --emit-moves > out.sam
# ont-guppy (tested with v6.3.7)
guppy_basecaller -c [DNA model] -i [INPUT] --moves_out --bam_out --save_path [OUTPUT]
samtools merge pass/*.bam -o pass_bam.bam # merge passed BAM files to create a single BAM file