def get_coding_potential_cutoff(ref_gtf, ref_fasta):
"""
estimate the coding potential cutoff that best classifies
coding/noncoding transcripts by splitting the reference
annotation into a test and training set and determining
the cutoff where the sensitivity and specificity meet
"""
train_gtf, test_gtf = gtf.split_gtf(ref_gtf, sample_size=2000)
coding_gtf = gtf.partition_gtf(train_gtf, coding=True)
noncoding_gtf = gtf.partition_gtf(train_gtf)
noncoding_fasta = gtf.gtf_to_fasta(noncoding_gtf, ref_fasta)
cds_fasta = gtf.gtf_to_fasta(coding_gtf, ref_fasta, cds=True)
hexamer_content = hexamer_table(cds_fasta, noncoding_fasta)
coding_fasta = gtf.gtf_to_fasta(coding_gtf, ref_fasta)
logit_model = make_logit_model(coding_fasta, noncoding_fasta,
hexamer_content, "test_gtf")
test_fasta = gtf.gtf_to_fasta(test_gtf, ref_fasta)
cpat_fn = cpat(test_fasta, hexamer_content, logit_model)
cpat_prob = load_cpat_coding_prob(cpat_fn)
coding, noncoding = gtf.get_coding_noncoding_transcript_ids(test_gtf)
best_score = 1
best_cutoff = 0
best_sensitivity = 0
best_specificity = 0
for cutoff in list(numpy.arange(0.1, 1, 0.01)):
grade = grade_cpat(coding, noncoding, cpat_prob, cutoff)
score = abs(grade["sensitivity"] - grade["specificity"])
if score < best_score:
best_score = score
best_cutoff = cutoff
best_sensitivity = grade["sensitivity"]
best_specificity = grade["specificity"]
return best_cutoff, hexamer_content, logit_model
def get_coding_potential_cutoff(ref_gtf, ref_fasta, data):
"""
estimate the coding potential cutoff that best classifies
coding/noncoding transcripts by splitting the reference
annotation into a test and training set and determining
the cutoff where the sensitivity and specificity meet
"""
train_gtf, test_gtf = gtf.split_gtf(ref_gtf, sample_size=2000)
coding_gtf = gtf.partition_gtf(train_gtf, coding=True)
noncoding_gtf = gtf.partition_gtf(train_gtf)
noncoding_fasta = gtf.gtf_to_fasta(noncoding_gtf, ref_fasta)
cds_fasta = gtf.gtf_to_fasta(coding_gtf, ref_fasta, cds=True)
hexamer_content = hexamer_table(cds_fasta, noncoding_fasta, data)
coding_fasta = gtf.gtf_to_fasta(coding_gtf, ref_fasta)
logit_model = make_logit_model(coding_fasta, noncoding_fasta,
hexamer_content, data, "test_gtf")
test_fasta = gtf.gtf_to_fasta(test_gtf, ref_fasta)
cpat_fn = cpat(test_fasta, hexamer_content, logit_model, data)
cpat_prob = load_cpat_coding_prob(cpat_fn)
coding, noncoding = gtf.get_coding_noncoding_transcript_ids(test_gtf)
best_score = 1
best_cutoff = 0
best_sensitivity = 0
best_specificity = 0
for cutoff in list(numpy.arange(0.1, 1, 0.01)):
grade = grade_cpat(coding, noncoding, cpat_prob, cutoff)
score = abs(grade["sensitivity"] - grade["specificity"])
if score < best_score:
best_score = score
best_cutoff = cutoff
best_sensitivity = grade["sensitivity"]
best_specificity = grade["specificity"]
return best_cutoff, hexamer_content, logit_model