search

tansey-lab / bayestme

BayesTME: A reference-free Bayesian method for analyzing spatial transcriptomics data
Other
14 stars 3 forks source link

question about runtime #148

Closed lllsssyyyy closed 6 months ago

jeffquinn-msk commented 7 months ago

hello~ I have been utilizing BayesTME for deconvolution analysis on a dataset containing 1000 spots and 20000 genes. I’ve noticed that the runtime extends beyond 4 hours, which seems quite lengthy. Could you kindly offer any guidance or suggestions to enhance the runtime efficiency?

Shirley

Were you using the (default) SVI inference mode (--inference-type SVI)? If so it should not take that long. Were you running the nextflow pipeline?

Best,

Jeff

lllsssyyyy commented 7 months ago

hello, Jeff After selecting the top 2000 genes, the runtime is shorter(using the default SVI mode), but is still around 30 minitues. Does BayesTME support gpu acceleration?

jeffquinn-msk commented 7 months ago

Hello,

Sorry were you not running the gene filtering step before deconvolution before? Things will not work correctly if you don't do that, we usually use top 1000 genes on our samples, pretty sure its going to be diminishing returns after that.

We don't support gpu acceleration yet, we will probably support that in the next version though.

Thanks,

Jeff