#!/usr/bin/env python3 """Genus-stratified subset of a SILVA build for comparing feature sets (k-mers vs Evo 2 ...). Takes up to --per sequences from each of --genera random genera that have at least --per sequences, so every sequence has same-genus neighbours to find. Writes OUT.fasta and OUT.tax.tsv (rows of the build's tax table). usage: pilot_sample.py nr99.tax.tsv SILVA_..._trunc.fasta.gz OUT [--genera 2000 --per 10 --seed 1] """ import argparse, gzip, random from collections import defaultdict ap = argparse.ArgumentParser() ap.add_argument("tax"); ap.add_argument("fasta"); ap.add_argument("out") ap.add_argument("--genera", type=int, default=2000); ap.add_argument("--per", type=int, default=10) ap.add_argument("--seed", type=int, default=1) a = ap.parse_args() lines = open(a.tax).read().splitlines() head, rows = lines[0], [l.split("\t") for l in lines[1:]] by = defaultdict(list) for r in rows: if r[7]: by[(r[1], r[7])].append(r) rng = random.Random(a.seed) ok = sorted(g for g, v in by.items() if len(v) >= a.per) pick = rng.sample(ok, min(a.genera, len(ok))) keep = {r[0]: r for g in pick for r in rng.sample(by[g], a.per)} with open(a.out + ".tax.tsv", "w") as t, open(a.out + ".fasta", "w") as o, gzip.open(a.fasta, "rt") as f: t.write(head + "\n") on = False for line in f: if line[0] == ">": sid = line[1:].split()[0] on = sid in keep if on: o.write(f">{sid}\n"); t.write("\t".join(keep[sid]) + "\n") elif on: o.write(line) print(f"{len(keep)} sequences from {len(pick)} of {len(ok)} genera with >= {a.per} -> {a.out}.fasta")