f5c-v1.2

No changes from v1.2-beta. Changes from v1.1 are as follows:

Major changes

• support for R10.4.1 flowcells (specify: --pore r10 option if FAST5, austodetected if S/BLOW5) for eventalign, call-methylation and resquiggle modules. See below for a benchmark.

Minor improvements

• sprintf_append dynamic (Sasha Jenner) #120
• prints stats on skipped reads due to low mapq, secondary mappings, etc. and warns when most reads are skipped due to lower mapq #122
• warn when more than half the reads fail to align/qc/calibration
• methylation models are unnecessarily not loaded for eventalign and resquiggle
• f4c3aaf: minor summary change
• 527ffc3: fix rsq

R10.4.1 f5c methylation calling benchmark

Correlation between whole genome NA12878 f5c CpG methylation calls (inhouse generated R10.4.1 PromethION data) and publicly available NA12878 bisulphite (no coverage filtering, that is, 1X also included):

Correlation between whole genome NA24385 f5c CpG methylation calls (inhouse generated R10.4.1 PromethION data) and publicly available NA24385 bisulphite data without any coverage filtering (no coverage filtering, that is, 1X also included):

The log-likelihood ratio for methylated vs unmethylated calls:

How R10.4.1 model training was done is detailed here. Anyone who have access to better training data may follow this tutorial to further improve the accuracy.