sample_posterior_predictive functionnot found

[Harrison-Q-Ma]

Greetings! I read your manuscript on bioRxiv and am trying to reproduce some of the analysis, specifically on estimating transcript diffusion. I was following the code in https://github.com/YosefLab/resolvi-reproducibility/blob/main/training_scripts/run_model.py but I got an error when I tried to run (line 84)

samples = model.sample_posterior_predictive(
        model=model.module.model_residuals,
        return_sites=[
            'mixture_proportions', 'mean_poisson', 'per_gene_background', 
            'diffusion_mixture_proportion', 'per_neighbor_diffusion', 'px_r_inv'
            ],
        batch_size=1000) 
samples = pd.DataFrame(samples).T 

It seems that sample_posterior_predictive is not a valid method in RESOLVI and I was not able to find it in the main scvi-tools github repo. I installed scvi-tools directly from the github repo and all other scvi.external.RESOLVI methods ran without error. Could you let me know how I can fix this error? Thanks in advance!

[canergen]

Thanks for reporting it. Indeed the reproducibility repository was run using the previous repository. The inference function was standardized with scVI-tools when merging the model (most other code remained the same, except changed between scVI-tools 1.0 and 1.2). The original model is at: https://github.com/canergen/resolVI. However, I would warn against using it except for strict reproducibility. The scVI-tools version is the maintained one. Please see our tutorial for the updated API called: model.sample_posterior (results are the same except for random seed differences).