Python API

The CLI is a thin wrapper. Both commands are one function call, so you can drive them from a pipeline without shelling out.

from genomic_benchmarks_qc.evaluate_classes import run as evaluate_classes
from genomic_benchmarks_qc.evaluate_splits import run as evaluate_splits

Full signatures and every argument are in the API reference.

Comparing classes

evaluate_classes(
    input=['examples/enhancers/data/enhancers_train.csv',
           'examples/enhancers/data/enhancers_test.csv'],
    format='csv',
    out_folder='qc-out',
)

format is required here — the CLI infers it from the file extension, but the function does not. It takes 'csv', 'tsv' or 'fasta'; gzip is detected per-file from the filename regardless.

Continuous labels, several sequence columns, and a trimmed per-position window:

evaluate_classes(
    input=['pairs.tsv'],
    format='tsv',
    sequence_column=['gene', 'noncodingRNA'],
    label_column='target',
    regression=True,
    end_position=500,
    out_folder='qc-out',
)

Checking a split

evaluate_splits(
    train_files=['train.csv'],
    test_files=['test.csv'],
    format='csv',
    sequence_column=['sequence'],
    similarity_threshold=90.0,
    out_folder='qc-out',
)

Note train_files / test_files rather than the CLI's --train-input / --test-input.

Reading the results back

Neither function returns the results — they write files, and the CSV is the interface. Reports land at <out_folder>/class/<column>/<classA>_vs_<classB>/ and <out_folder>/split/<column>/<train>_vs_<test>/, with directory names derived from your labels and column names.

Rather than reconstruct those paths, glob for the report:

import csv, pathlib

for report in pathlib.Path('qc-out').rglob('gb-qc-report.csv'):
    with report.open(newline='') as handle:
        flags = {r['Check']: r['Flag'] for r in csv.DictReader(handle)
                 if ' - ' not in r['Check']}          # headline checks only
    failed = [c for c, f in flags.items() if f == 'Fail']
    print(report.parent.name, failed or 'clean')

Adding 'json' to report_types also writes per-class statistics — counts, GC, lengths, base and dinucleotide frequencies — to <out_folder>/class/<column>/per-class/<class>.json, which is the machine-readable route to the numbers behind the flags.

Logging and failures

Both functions call setup_logger and log progress at INFO. Pass log_level='WARNING' for quiet, or log_file='qc.log' to tee to a file.

Only the genomic_benchmarks_qc logger is configured — your root logger, and every other library reporting through it, is left as you set it up. Calling either function again reconfigures that logger, so two runs in one process each get the level and the file they asked for. To route the output yourself instead, ignore both arguments and attach a handler:

import logging

logging.getLogger('genomic_benchmarks_qc').addHandler(your_handler)

They raise on failure rather than exiting, so wrap them as you would any other call. The CLI's non-zero exit codes are its own translation layer, not the functions' behaviour.