Missing RNA edit calls

[physnano]

I am attempting to run Clair3-rna on PacBio data (3 biological replicates of a sample) to identify RNA editing sites using the following:

PLATFORM='hifi_mas_pbmm2'
MIN_COV=10
${BASE}run_clair3_rna \
    --bam_fn ${BAMIN} \
    --ref_fn ${REF} \
    --threads ${THREADS} \
    --platform ${PLATFORM} \
    --tag_variant_using_readiportal \
    --min_coverage ${MIN_COV} \
    --sample_name=${SAMPLE} \
    --output_dir ${OUTPUT_DIR}

Based on a known RNA editing site and visual inspection of IGV reads (See below) I should be seeing the edit in all three replicates of this sample. However only one of the three replicate output.vcf.gz files actually calls the edit site despite ample coverage in the other replicates. What could be the cause of this discrepancy?

Clair3-rna VCF output:

IGV:

[zhengzhenxian]

Hi @physnano,

Clair3-RNA does not classify RNA editing as a variant; instead, we trained all editing sites as non-variants. As a result, you may notice that most editing events are either missing or labeled as "RefCall" in the output VCF. And for those reliable editing sites collected in REDIportal and reported as variants in output VCF, we tag them as "RNAEditing".

Please consider using RNA editing detection tools such as Modkit for identifying RNA editing.

[physnano]

Hi @zhengzhenxian

I understand that it is an RNAEditing site and will be tagged as such. My question relates to why it is detected in one sample (r2) but not r1 or r3 when clearly there is obvious RNA editing event (with sufficient coverage) as indicated by visual inspection of reads in IGV. What could explain this discrepancy?