Running SpliceAI takes too much time

[gswsqffsapd3056]

When I annotate a 30,000-site mutant vcf using SpliceAI, it took almost two hours to complete the entire annotation process
And I got a Warning from Tensorflow WARNING:tensorflow:5 out of the last 395 calls to <function Model.make_predict_function.<locals>.predict_function> triggered tf.function retracing. Tracing is expensive ...
I don't know if this situation is related to the warning message


I use a Quadro RTX 6000 GPU model CUDA Version: 11.7

[h-joshi]

Hi Juliette @gswsqffsapd3056 - Could you please try using this build?
https://github.com/kidsneuro-lab/SpliceAI/releases/tag/v1.3.4

Please refer to the README

• The VCF output has additional fields to capture the raw REF/ALT scores
• The grch38_custom annotation is built from Canonical/MANE transcripts derived from Ensembl 104 / Gencode 38
• The model loading is optimised. This should in theory speed up the inference

Would be great to know how you've fared with this updated build.