how to create negative samples?
wawpaopao opened this issue · 1 comments
"To generate a single negative sample, a random reference bacterium and a random sample from the positive ClusterFinder set were selected. Each gene in the positive sample was replaced with a random gene from the reference bacteria, while considering only 1% of genes that were most similar in number of Pfam domains. In total, three samples were generated from each reference bacteria, producing 10 128 negative samples."
In theory, a lot of negative samples can be created. Have you tried to generate other numbers of negative samples for training?
And I don't figure out how to create a negative sample. In the above description, the operation object seems to be positive sample instead of reference bacteria.After removing the regions similar to MIBiG,How did you create negative samples？
Thank you very much!
Hi @wawpaopao we haven't experimented with other settings or other methods of generating negative sets. You can download our negative samples "GeneSwap_Negatives.pfam.tsv" in format supported for DeepBGC training from this release page: https://github.com/Merck/deepbgc/releases/tag/v0.1.0