--seed 0 silently picks a random seed — print or pin the effective seed #154

protein_mpnn_run.py treats --seed 0 (the default) as a request for a random seed:

if args.seed:
    seed=args.seed
else:
    seed=int(np.random.randint(0, high=999, size=1, dtype=int)[0])

The help text does say "If set to 0 then a random seed will be picked", so this is intentional — but it's a footgun for downstream callers: passing --seed 0 expecting determinism (a natural reading, since 0 is a valid seed in most tools) silently produces different sequences on every run. We hit this in a pipeline wrapper where seed: 0 in config was assumed to pin sampling; results were irreproducible until we noticed the falsy check.

Two possible improvements:

1. Print the effective seed at startup — it is already recorded in output fasta headers (seed={}), but that is post-hoc; a one-line print makes randomized runs auditable as they launch.
2. Or accept 0 as a literal seed — e.g. default None and seed = args.seed if args.seed is not None else random_draw(), so an explicit --seed 0 is deterministic and only an absent flag randomizes.

Happy to open a PR for whichever direction you prefer. Our downstream fork currently does option 1 (prints the effective seed when 0 triggers a random draw).