def classify_data_by_model(
postprocessed_ions_file, model_file_path, method="full_random_forest", out=None, method_init_args=None,
method_fit_args=None):
task = ClassifyTargetWithModelTask(
model_file_path, postprocessed_ions_file, method=method, output_path=out,
method_init_args=method_init_args, method_fit_args=method_fit_args)
result = task.run()
return result
def main(scored_matches_path, decon_data=None, model_file_path=None, decoy_matches_path=None,
outfile_path=None, num_decoys_per_real_mass=1.0, random_only=False,
predicate_fns=None, prefix_len=0, suffix_len=0, by_mod_sig=False,
ms1_tolerance=None, ms2_tolerance=None, method="full_random_forest", method_init_args=None,
method_fit_args=None, n_processes=6):
'''
Call with deconvolution results and a model to generate decoys and score them, or with
a pre-existing decoy database.
:type predicate_fns: Sequence
:param predicate_fns iterable: containing functions with which to partition both the
"real" and "decoy" databases. Use `make_predicate` with keyword arguments matching column
names and numeric thresholds for ease of use and documentation.
:param outfile_path str: defaults to scored_matches_path[:-4] + "_fdr.json" will contain the
resulting FDR statistic for each cutoff.
'''
scored_matches_frame = classify_matches.prepare_model_file(scored_matches_path)
decoy_matches_frame = None
if ms1_tolerance is None:
ms1_tolerance = scored_matches_frame.metadata["ms1_ppm_tolerance"]
if ms2_tolerance is None:
ms2_tolerance = scored_matches_frame.metadata["ms2_ppm_tolerance"]
if outfile_path is None:
outfile_path = scored_matches_path[:-5] + "_fdr.json"
if decon_data is not None and model_file_path is not None:
logger.info("No decoys given.")
decoy_matches_path = generate_decoy_match_results(
scored_matches_path, decon_data, model_file_path, prefix_len=prefix_len,
suffix_len=suffix_len, ms1_tolerance=ms1_tolerance, ms2_tolerance=ms2_tolerance,
num_decoys_per_real_mass=num_decoys_per_real_mass, random_only=random_only,
method=method, method_init_args=method_init_args, method_fit_args=method_fit_args,
n_processes=n_processes, outfile_path=outfile_path)
decoy_matches_frame = classify_matches.prepare_model_file(
decoy_matches_path)
elif model_file_path is not None and decoy_matches_path is not None:
scored_matches_frame = ClassifyTargetWithModelTask(
model_file_path, scored_matches_path, method=method).run(False)
decoy_matches_frame = ClassifyTargetWithModelTask(
model_file_path, decoy_matches_path, method=method).run(False)
else:
logger.info("Pre-matched Decoys Given")
scored_matches_frame = classify_matches.prepare_model_file(
scored_matches_path)
decoy_matches_frame = classify_matches.prepare_model_file(
decoy_matches_path)
logger.info("Evaluating predicates")
fdr_search = optimize_fdr.CountExclusion(
scored_matches_frame, decoy_matches_frame,
decoy_matches_frame.metadata["decoy_ratio"], ["MS2_Score"])
fdr_search.optimize()
scored_matches_frame.metadata["fdr"] = fdr_search.compress()
scored_matches_frame["call"] = scored_matches_frame.Glycopeptide_identifier.isin(
scored_matches_frame.kvquery().Glycopeptide_identifier)
try:
logger.info("Accepted %d predictions with FDR %f",
scored_matches_frame.call.sum(),
scored_matches_frame.optimize_fdr().iloc[0]["false_discovery_rate"])
except:
logger.info("No predictions were found at an acceptable FDR")
scored_matches_frame.serialize(outfile_path)
return outfile_path