def adjudicate_BAF(metrics, labeler, name):
# Deletions
testable = metrics.loc[(metrics.svtype == 'DEL')
& (metrics.svsize >= 5000)]
trainable = testable.loc[(testable.poor_region_cov < 0.3)
& ~testable.chrom.isin(ALLOSOMES)
& ~testable.is_outlier_specific]
trainable['label'] = labeler.label(trainable)
trainable.to_csv('{0}_DEL_trainable.txt'.format(name),
index=False,
sep='\t')
testable.to_csv('{0}_DEL_testable.txt'.format(name), index=False, sep='\t')
features = 'BAF_snp_ratio BAF_del_loglik'.split()
cutoffs = {'indep': ['BAF_snp_ratio'], 'dep': ['BAF_del_loglik']}
del_cutoffs = rf_classify(metrics, trainable, testable, features, labeler,
cutoffs, name)
# Duplications
testable = metrics.loc[(metrics.svtype == 'DUP')
& (metrics.svsize >= 5000)]
trainable = testable.loc[(testable.poor_region_cov < 0.3)
& ~testable.chrom.isin(ALLOSOMES)
& ~testable.is_outlier_specific]
trainable['label'] = labeler.label(trainable)
trainable.to_csv('{0}_DUP_trainable.txt'.format(name),
index=False,
sep='\t')
testable.to_csv('{0}_DUP_testable.txt'.format(name), index=False, sep='\t')
features = 'BAF_KS_stat BAF_KS_log_pval'.split()
cutoffs = {'indep': ['BAF_KS_stat'], 'dep': ['BAF_KS_log_pval']}
dup_cutoffs = rf_classify(metrics, trainable, testable, features, labeler,
cutoffs, name)
# Combine cutoffs
del_cutoffs['svtype'] = 'DEL'
dup_cutoffs['svtype'] = 'DUP'
cutoffs = pd.concat([del_cutoffs, dup_cutoffs]).reset_index()
cutoffs.to_csv('{0}_cutoffs.txt'.format(name), index=False, sep='\t')
cutoffs['test'] = 'BAF'
cutoffs['max_svsize'] = np.nan
cutoffs['min_svsize'] = 5000
cutoffs['algtype'] = 'any'
return cutoffs
def adjudicate_SR1(metrics):
testable = metrics.loc[~metrics.name.str.contains('_depth_')]
trainable = testable.loc[(testable.poor_region_cov < 0.3)
& ~testable.chrom.isin(ALLOSOMES)
& ~testable.is_outlier_specific]
features = ['SR_sum_log_pval', 'SR_sum_bg_frac']
cutoffs = {'indep': ['SR_sum_log_pval'], 'dep': ['SR_sum_bg_frac']}
labeler = labelers.SR1TrainingLabeler()
trainable['label'] = labeler.label(trainable)
trainable.to_csv('SR1_trainable.txt', index=False, sep='\t')
cutoffs = rf_classify(metrics, trainable, testable, features, labeler,
cutoffs, 'SR1_prob')
cutoffs.to_csv('SR1_cutoffs.txt', index=False, sep='\t')
cutoffs['test'] = 'SR1'
cutoffs['svtype'] = 'CNV'
cutoffs['algtype'] = 'PESR'
return cutoffs
def adjudicate_RD(metrics):
features = ["RD_Median_Separation", "RD_log_pval", "RD_log_2ndMaxP"]
cutoff_features = {
'indep': ['RD_log_pval', 'RD_Median_Separation'],
'dep': ['RD_log_2ndMaxP']
}
labeler = labelers.RDTrainingLabeler()
cutoff_dfs = []
# PE/SR >1 kb
testable = metrics.loc[~metrics.name.str.contains('_depth_')
& (metrics.svsize >= 1000)]
trainable = testable.loc[(testable.svsize >= 5000)
& (testable.poor_region_cov < 0.3)
& ~testable.chrom.isin(ALLOSOMES)
& ~testable.is_outlier_specific]
testable.to_csv('RD_pesr_gt5kb_testable.txt', index=False, sep='\t')
trainable['label'] = labeler.label(trainable)
trainable.to_csv('RD_pesr_gt5kb_trainable.txt', index=False, sep='\t')
cutoffs = rf_classify(metrics, trainable, testable, features, labeler,
cutoff_features, 'RD_prob')
cutoff_dfs.append(cutoffs)
cutoff_dfs[0]['algtype'] = 'PESR'
cutoff_dfs[0]['max_svsize'] = np.nan
cutoff_dfs[0]['min_svsize'] = 1000
cutoff_dfs[0]['svtype'] = 'CNV'
# PE/SR <1 kb
testable = metrics.loc[~metrics.name.str.contains('_depth_')
& (metrics.svsize < 1000)]
trainable = testable.loc[(testable.svsize >= 100)
& (testable.poor_region_cov < 0.3)
& ~testable.chrom.isin(ALLOSOMES)
& ~testable.is_outlier_specific]
testable.to_csv('RD_pesr_lt5kb_testable.txt', index=False, sep='\t')
trainable['label'] = labeler.label(trainable)
trainable.to_csv('RD_pesr_lt5kb_trainable.txt', index=False, sep='\t')
cutoffs = rf_classify(metrics, trainable, testable, features, labeler,
cutoff_features, 'RD_prob')
cutoff_dfs.append(cutoffs)
cutoff_dfs[1]['algtype'] = 'PESR'
cutoff_dfs[1]['max_svsize'] = 1000
cutoff_dfs[1]['min_svsize'] = 0
cutoff_dfs[1]['svtype'] = 'CNV'
# Depth dels
cutoff_features = {
'indep': ['RD_log_pval', 'RD_Median_Separation'],
'dep': []
}
testable = metrics.loc[metrics.name.str.contains('_depth_')
& (metrics.svtype == 'DEL')]
trainable = testable.loc[(testable.svsize >= 5000)
& (testable.poor_region_cov < 0.3)
& ~testable.chrom.isin(ALLOSOMES)
& ~testable.is_outlier_specific]
trainable['label'] = labeler.label(trainable)
trainable.to_csv('RD_depth_DEL_trainable.txt', index=False, sep='\t')
cutoffs = rf_classify(metrics,
trainable,
testable,
features,
labeler,
cutoff_features,
'RD_prob',
clean_cutoffs=True)
cutoff_dfs.append(cutoffs)
cutoff_dfs[2]['algtype'] = 'Depth'
cutoff_dfs[2]['max_svsize'] = np.nan
cutoff_dfs[2]['min_svsize'] = 5000
cutoff_dfs[2]['svtype'] = 'DEL'
# Depth dups
testable = metrics.loc[metrics.name.str.contains('_depth_')
& (metrics.svtype == 'DUP')]
trainable = testable.loc[(testable.svsize >= 5000)
& (testable.poor_region_cov < 0.3)
& ~testable.chrom.isin(ALLOSOMES)
& ~testable.is_outlier_specific]
trainable['label'] = labeler.label(trainable)
trainable.to_csv('RD_depth_DUP_trainable.txt', index=False, sep='\t')
cutoffs = rf_classify(metrics,
trainable,
testable,
features,
labeler,
cutoff_features,
'RD_prob',
clean_cutoffs=True)
cutoff_dfs.append(cutoffs)
cutoff_dfs[3]['algtype'] = 'Depth'
cutoff_dfs[3]['max_svsize'] = np.nan
cutoff_dfs[3]['min_svsize'] = 5000
cutoff_dfs[3]['svtype'] = 'DUP'
# Fail depth-only below 5 kb
metrics.loc[metrics.name.str.contains('_depth_') & (metrics.svsize < 5000)
& (metrics.RD_prob >= 0.5), 'RD_prob'] = 0.499
cutoffs = pd.concat(cutoff_dfs)
cutoffs.to_csv('RD_cutoffs.txt', index=False, sep='\t')
cutoffs['test'] = 'RD'
return cutoffs