Fix: include sample-selection parameters in PLINK output filename

[Mowblow]

Fix: include sample-selection parameters in PLINK output filename

Fixes #1113

Problem

The filename prefix for biallelic_snps_to_plink was built from only a subset of input parameters:

plink_file_path = f"{output_dir}/{region}.{n_snps}.{min_minor_ac}.{max_missing_an}.{thin_offset}"

Parameters like sample_sets, sample_query, sample_query_options, sample_indices, site_mask, and random_seed were excluded, meaning two calls with different sample cohorts could silently resolve to the same filename and overwrite each other.

Approach

I considered three options:

1. Append raw parameter values to the filename — rejected because sample_sets can be a list and sample_query can be a long arbitrary string, making filenames unpredictable and potentially invalid on some operating systems.

2. Include only the most commonly varied parameters — rejected because it still leaves some parameters out and the same collision problem could reappear.

3. Hash all sample-selection parameters together — chosen because it keeps filenames short and OS-safe, guarantees uniqueness across any combination of inputs, and handles None values and lists cleanly through JSON serialization.

Solution

Bundle all sample-selection parameters into a dictionary, serialize with json.dumps (using sort_keys=True for consistency and default=str for safety), and append an 8-character MD5 hash to the filename:

sample_params = {
    "sample_sets": sample_sets,
    "sample_query": sample_query,
    "sample_query_options": sample_query_options,
    "sample_indices": sample_indices,
    "site_mask": site_mask,
    "random_seed": random_seed,
}
params_hash = hashlib.md5(
    json.dumps(sample_params, sort_keys=True, default=str).encode()
).hexdigest()[:8]

plink_file_path = f"{output_dir}/{region}.{n_snps}.{min_minor_ac}.{max_missing_an}.{thin_offset}.{params_hash}"

The resulting filename is clean, stable, and unique per parameter combination — for example: 2L.1000.2.8.0.a3f7bc12.

Testing

I verified that:

• Two calls with different sample_sets produce different filename prefixes
• Two calls with identical parameters produce the same filename prefix (caching still works)
• None values are handled without errors