Multiple Errors on Mac Silicon

[BMTTMC]

I'm trying to run variant calling on ONT data using my MacBook Pro using Apple Silicon (M2). When using tools such as freebayes and Clair3 I get multiple variants, but I'm running into issues when trying to run ClairS-TO. The script finishes without finding any variants and my concern is that the reason is caused by some of the errors I see in the logs:

[INFO] STEP 2: SNV Pileup Model Calling
...
[INFO] SNV Pileup Model Prediction along with Affirmative & Negational Model
...
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr18 total processed positions: 4, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr19 total processed positions: 1, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr7 total processed positions: 4, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr12 total processed positions: 2, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr15 total processed positions: 3, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr17 total processed positions: 6, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr4 total processed positions: 1, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr19 total processed positions: 11, time elapsed: 1.2s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr17 total processed positions: 1, time elapsed: 0.7s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr1 total processed positions: 3, time elapsed: 0.8s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr12 total processed positions: 1, time elapsed: 0.7s
[W NNPACK.cpp:61] Could not initialize NNPACK! Reason: Unsupported hardware.
[INFO] chr12 total processed positions: 2, time elapsed: 0.8s

While this could be a red herrings, as indicated by the continuation of the script, further errors appear later:

[INFO] STEP 3: Non-somatic Tagging for SNV Variants
...

cWarning: cannot find your CPU L2 cache size in /proc/cpuinfo

gzip: stdout: Broken pipe
Killed
Warning: cannot find your CPU L2 cache size in /proc/cpuinfo

These warnings continue throughout multiple future steps. This is the error that concerns me the most. Any suggestions on how to fix these? Or are they just warnings that can be ignored? Thanks!

[aquaskyline]

Hi Ben, are you running ClairS-TO natively on Apple Silicon, or through docker virtualization?

@JasonCLEI consider disabling NNPACK when MacOS is detected os.environ["TORCH_NNPACK"] = "0".

[JasonCLEI]

Hi @BMTTMC,

I'm testing ClairS-TO on Apple Silicon. But through your partial running log, I think the issue is caused by other problems besides the Mac Silicon. Could you please provide your complete running log of run_clairs_to.log file with us, and we will check the issue for you further.

Lei

[BMTTMC]

Thank you @aquaskyline and @JasonCLEI .
I'm running on docker using the following command:
docker run -i -v {INPUT}:{INPUT} -v {OUTPUT}:{OUTPUT} hkubal/clairs-to:latest /opt/bin/run_clairs_to --tumor_bam_fn {INPUT_BAM} --ref_fn {GENOME_REF} --thread 8 --platform ont_r10_dorado_hac_4khz --output_dir {OUTPUT}

I'm attaching the run_clairs_to_log as requested.

run_clairs_to.log

Thank you for your guidance!

[JasonCLEI]

Hi @BMTTMC,

After testing ClairS-TO on Apple Silicon, we found that there are indeed some hardware architecture compatibility issues on Apple Silicon M1/M2 with ClairS-TO. And we are trying to adapt ClairS-TO to Apple Silicon M1/M2. However, these compatibility warnings will not result in an obvious error. Through your complete running log, I think the issue is caused by other problems besides the Apple Silicon M1/M2. If you want to explore the issue further, you can send me a mini bam of your data and other necessary materials via my email (lchen@cs.hku.hk), and we could help check the issue for you further.

Lei