ClairS 0.4.3 generating empty snp and indel vcfs on HCC1395 test

[insilicool]

Hi,

I’m currently testing ClairS 0.4.3 (conda build) on the HCC1395 benchmark ONT dataset from your paper (SRR25005626). However, when I run tests on both chr1 and chrM, the output contains empty SNV and INDEL files.

Here is the command I’m using:

module purge && source ~/.bashrc && micromamba activate clairs_043 &&
${CONDA_PREFIX}/bin/ClairS/run_clairs
--enable_indel_calling --print_germline_calls --enable_clair3_germline_output
--platform ont_r10_dorado_sup_5khz
--clair3_model_path="${MUGQIC_INSTALL_HOME}/genomes/rerio/clair3_models/r1041_e82_400bps_sup_v430"
--threads 5 --output_dir clairS_043/HCC1395/chr1
--ctg_name chr1
--ref_fn ${MUGQIC_INSTALL_HOME_DEV}/genomes/species/Homo_sapiens.GRCh38/genome/Homo_sapiens.GRCh38.fa
--sample_name HCC1395
--conda_prefix ${CONDA_PREFIX}
--normal_bam_fn alignment/HCC1395BL/HCC1395BL.sorted.bam
--tumor_bam_fn alignment/HCC1395/HCC1395.sorted.bam

Running the same command with ClairS 0.3.0 produces results as expected. I’ve attached the log file for reference.

Could you please advise on what might be causing the issue with version 0.4.3?

Thank you for your time,
Rob

run_clairs.log

[insilicool]

As a follow-up to my previous message:

I built a Singularity container for ClairS 0.4.3, and after some adjustments to the command (shown below), I was able to successfully generate SNV and INDEL variants for chr 1:

singularity exec
clairs_043.sif
bash -c "export LC_ALL=C.UTF-8 LANG=C.UTF-8 &&
source /opt/conda/etc/profile.d/conda.sh &&
conda activate clairs &&
/opt/bin/run_clairs

The command executed correctly inside the container, and the output included variants as expected.

One notable difference I observed in the logs between the conda build and the container build is the Clair3 version:

ClairS 0.4.3 (conda): [INFO] CLAIR3 VERSION: v1.0.10

ClairS 0.4.3 (container): [INFO] CLAIR3 VERSION: v1.0.7

Could this discrepancy indicate that the conda environment setup is incorrect or missing something?

Please advise.

Thank you for your time

[zhengzhenxian]

Hi @insilicool,

It is expected. The singularity image is built using Docker, with the previous Clair3 version utilized in the Docker build cache few months ago. In contrast, Bioconda will automatically seek the latest version of Clair3 for installation. It should be all good for using two Clair3 versions in ClairS workflow.

[insilicool]

Hi @zhengzhenxian,

Thank you for your response.

Would this imply that the conda version could produce different results compared to the Singularity/Docker version due to the difference in Clair3 versions?

That said, it still doesn’t fully explain why the ClairS conda build produces empty SNV and INDEL VCFs for me, while the ClairS Singularity container (with the same input) generates non-empty VCFs for the same chromosome.

I suspect the issue might be related to the conda installation workflow. For example, the following command did not work as expected for me:
conda create -n clairs_043 -c bioconda -c pytorch -c conda-forge pytorch tqdm clair3-illumina python=3.9.0 -y

However, specifying the Clair3 version to match the one in the Singularity image (v1.0.7) did:
conda create -n clairs_test_043 -c bioconda -c pytorch -c conda-forge pytorch tqdm clair3=1.0.7 python=3.9.0 -y