Different Bin Counts

[beaPhD]

Hello, I have a question about running VAMB. I'm not using the mixed assembly approach, but rather doing single-sample assembly. Therefore, I didn't run ./vamb/src/concatenate.py initially and directly used the MEGAHIT assembly results as input. In this case, VAMB generated 49 bins.

However, when I ran ./vamb/src/concatenate.py with the --keepnames parameter (to not modify sequence names) and filtered out sequences shorter than 2000bp, then used this as input to VAMB, it only generated 48 bins.

I'm confused because in both cases, I used the -m 2000 parameter when running VAMB, so theoretically the results should be the same. Could you please explain why there's this discrepancy in the number of bins generated?

[jakobnissen]

Vamb is not deterministic unless the seed is set (and even in the presence of a seed, we cannot guarantee determinism because PyTorch doesn't).
Did you set the random seed to the same?

[beaPhD]

Thank you for your answer. After setting seed=12345 and running twice on the same sample, I found that the number of bins was consistent and the sequences within the bins were identical (only the sequence order differed). I would like to ask if you have any recommended random seed numbers when developing software? I plan to use VAMB for single-sample rapid binning.

[jakobnissen]

There is no right or wrong seed: All seeds are interchangeable. When developing Vamb itself, it's best to not set the seed (since otherwise, random fluctuations may more easily be mistaken for algorithmic improvements). When developing other software using Vamb, set the seed to a fixed number, but the number itself doesn't matter.