Random seed in AlphaFold

Question

[aaggencc]

Hi, I am looking at AlphaFold versions in google notebook but there is the expression of random seed somewhere. Actually what does this mean in protein modeling in AlphaFold?I would be very appreciative to introduce about this from someone who knows.

Answer 1

[Wayne]

Practical concerns regard in the available Google Colab notebooks:

This guide posted at Proteopedia mentions the seed.

"num_samples (random seeds): Leave this at 1. Beware that if you increase this above 1, you will generate a number of models equal to the product of this value times num_models. This will proportionally increase the time to complete a result."

I've also noted this post from August 2021:

"When there are few to no sequences in the MSA, we find sometimes changing the random_seed, or running for more recycles allows #alphafold to predict the correct structure. Notebook if anyone wants to try on their difficult cases: https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/beta/AlphaFold2_advanced.ipynb "

---

From a computational aspect, the use of different random seeds adds the possibility of variance in the prediction outcomes.

"Because of the random seeds used in AlphaFold, and also because of processes like GPU inference that are nondeterministic, the prediction results in AlphaFold ... may have run variance." - SOURCE: Zhong et al. 'ParaFold: Paralleling AlphaFold for Large-Scale Predictions'

For computer generation of a series of 'random numbers', a seed value is first provided to kick off the process. The series produced subsequently is meant to approximate randomness, often called pseudo-random number generation. Because it actually is deterministic, using the same seed generates the same stream of pseudo-random numbers. Switching seeds you can get a different stream of pseudo-random numbers, and thus different subsequent calculations & results done with derivatives of that stream.