Own hg38-fasta usage fails

[sanjansen]

Dear Team decima,

Our names are Sanne Jansen and Orfeas Gkourlias, both student of the Master Data Science for Life Sciences followed at the Hanze University of Applied Sciences in Groningen (Netherlands).

Currently, we are working on a project for which we would like to deploy your model, Decima. Specifically for the prediction of the effects of variants (predict_variant_effect), using your Python API. Therefore, we have installed it on the cluster, and have successfully performed our first test-run.
However, when we supply our own hg38 fasta-file within the ''genome = '' argument, it fails to use this specified hg38 file and still downloads the hg38-file from wandb.

We were wondering whether this is a known issue, and if so how we could solve it for Decima to use our own supplied hg38-fasta file.

The version of the operating system: Linux Rocky version 9.5
The version of Python: Python 3.13.5.

Thank you for your time in advance, and we hope to hear from you soon.

[MuhammedHasan]

Hi @sanjansen,

Can you provide the code you run and the output you get? What is your decima and grelu version?

[sanjansen]

Hi @MuhammedHasan, thank you for your quick response!

The code I run:

def main():
'''
Main function...
'''
df_variant = pd.read_table("/path/to/chr2_test_decima.tsv")
predict_variant_effect(df_variant,
output_pq = "/path/to/vep_results_chr2_cs_py.parquet" ,
device = device,
tasks = "cell_type == 'B cell' ",
genome = "/path/to/hg38/hg38.fa")

In the output I get a parquet file as expected (also shown in your documentation).

Versions
decima==0.3.3

However, we did not use deploy gReLU to run Decima, as I have not seen it mentioned in the documentation. Perhaps there could be an issue?

Kind regards,

Sanne and Orfeas

[MuhammedHasan]

Can you share your grelu version as well with following command? (https://github.com/Genentech/gReLU)

pip freeze | grep gReLU

[sanjansen]

My apologies, I thought I had done my pip freeze correctly before. But it turns out I did not. Here is indeed the gReLu version:
gReLU==1.0.10

[MuhammedHasan]

How do you know it did not used your fasta file? I think it worked as expected and used the fasta file your provided.

[sanjansen]

When I run the program, I can see in the output log file that it downloads the FASTA file from somewhere else. Sadly, I cannot show the output log file as of right now, as I am not able to get to the file at this specific time.

[MuhammedHasan]

Hi @sanjansen,

I updated code base to fix the issue reported above with recent PR #40.

Can you try again and report if issue is still present?

[sanjansen]

Thank you for your help, it indeed looks like it does not download the HG38 from wandb anymore, so it seems like the problem has been fixed.