141 lines (129 with data), 6.5 kB

include: 'common.snakemake'

import os
import yaml
with open(data_dir + '/compare_groups.yaml', 'r') as f:
    compare_groups = yaml.load(f)
with open(get_config_file('evaluate_features.yaml'), 'r') as f:
    cv_config = yaml.load(f)

classifiers = list(cv_config['classifiers'].keys())

inputs = {'evaluate_features': []}
for compare_group, feature_set in get_known_biomarkers():
    inputs['evaluate_features'] += expand('{output_dir}/evaluate_features/{compare_group}/{feature_set}/filter.{imputation_method}.Norm_{normalization_method}.Batch_{batch_removal_method}_{batch_index}.{count_method}/{classifier}',
        output_dir=output_dir, 
        imputation_method=config['imputation_method'],
        normalization_method=config['normalization_method'],
        batch_removal_method=config['batch_removal_method'],
        batch_index=config['batch_index'],
        count_method=config['count_method'],
        classifier=classifiers, 
        compare_group=compare_group,
        feature_set=feature_set)
inputs['summarize_evaluate_features'] = expand('{output_dir}/summary/evaluate_features/{summary_name}.txt',
    output_dir=output_dir, summary_name=['metrics.train', 'metrics.test'])


rule all:
    input:
        unpack(lambda wildcards: inputs)


rule preprocess_features:
    input:
        '{output_dir}/evaluate_features/matrix/{compare_group}/{feature_set}.txt'
    output:
        '{output_dir}/evaluate_features/preprocess_features/{compare_group}/{feature_set}.txt'
    params:
        scaler=config['scale_method']
    shell:
        '''{bin_dir}/feature_selection.py preprocess_features -i {input} --scaler {params.scaler} \
            --use-log --transpose -o {output}
        '''

rule evaluate_features:
    input:
        matrix='{output_dir}/matrix_processing/{preprocess_method}.{count_method}.txt',
        sample_classes=data_dir+ '/sample_classes.txt',
        features=data_dir + '/known_biomarkers/{compare_group}/{featureset}.txt'
    output:
        dir=directory('{output_dir}/evaluate_features/{compare_group}/{featureset}/{preprocess_method}.{count_method}/{classifier}')
    run:
        from copy import deepcopy

        output_config = {}
        # copy global config parameters
        for key in ('transpose', 'features', 'cv_params', 'sample_weight', 'preprocess_steps'):
            if key in cv_config:
                output_config[key] = cv_config[key]
        # copy classifier config
        classifier_config = deepcopy(cv_config['classifiers'][wildcards.classifier])
        classifier_config['params'] = classifier_config.get('params', {})
        output_config['classifier'] = classifier_config['classifier']
        output_config['classifier_params'] = classifier_config.get('classifier_params', {})
        # copy classifier grid search params
        if classifier_config.get('grid_search', False):
            grid_search_params = deepcopy(cv_config['classifier_grid_search_params'])
            grid_search_params.update(classifier_config['grid_search_params'])
            # add classifier grid search config
            output_config['grid_search'] = True
            output_config['grid_search_params'] = grid_search_params
        # write output config
        if not os.path.isdir(output.dir):
            os.makedirs(output.dir)
        output_config_file = os.path.join(output.dir, 'config.yaml')
        with open(output_config_file, 'w') as f:
            yaml.dump(output_config, f, default_flow_style=False)
        command = [
            os.path.join(config['bin_dir'], 'machine_learning.py'), 'run_pipeline',
            '--matrix', input.matrix,
            '--sample-classes', input.sample_classes,
            '--output-dir', output.dir,
            '--features', input.features,
            '--positive-class', '"' + compare_groups[wildcards.compare_group][1] + '"',
            '--negative-class', '"' + compare_groups[wildcards.compare_group][0] + '"',
            '--config', output_config_file
        ]
        shell(' '.join(command))

"""
rule evaluate_features:
    input:
        matrix='{output_dir}/matrix_processing/filter.{imputation_method}.Norm_{normalization_method}.Batch_{batch_removal_method}_{batch_index}.{count_method}.txt',
        sample_classes=data_dir+ '/sample_classes.txt',
        features=data_dir + '/known_biomarkers/{compare_group}/{feature_set}.txt'
    output:
        directory('{output_dir}/evaluate_features/{compare_group}/{feature_set}/filter.{imputation_method}.Norm_{normalization_method}.Batch_{batch_removal_method}_{batch_index}.{count_method}/{classifier}')
    params:
        count_method=count_method_regex
    run:
        import json
        import os
        import subprocess
        from shlex import quote
        from copy import deepcopy

        command = [
            os.path.join(config['bin_dir'], 'machine_learning.py'), 'cross_validation',
            '--matrix', input.matrix,
            '--sample-classes', input.sample_classes,
            '--output-dir', output[0],
            '--transpose',
            '--positive-class', compare_groups[wildcards.compare_group][1],
            '--negative-class', compare_groups[wildcards.compare_group][0],
            '--cv-params', json.dumps(config['cv_params']),
            '--selector', 'null',
            '--features', input.features
        ]
        if config['log_transform']:
            command += ['--log-transform', '--log-transform-params', json.dumps(config['log_transform_params'])]
        if config['scaler']:
            command += ['--scaler', config['scaler'], '--scaler-params', json.dumps(config['scaler_params'].get(config['scaler'], {}))]
        #if config['grid_search']:
        #    command += ['--grid-search', '--grid-search-params', json.dumps(config['grid_search_params'])]
        if config['sample_weight']:
            command += ['--sample-weight', config['sample_weight']]
        command += ['--classifier', wildcards.classifier, 
            '--classifier-params', json.dumps(config['classifier_params'].get(wildcards.classifier, {}))]
        command = list(map(str, command))
        print(' '.join(map(quote, command)))
        subprocess.check_call(command)
"""

rule summarize_evaluate_features:
    input:
        input_dir=inputs['evaluate_features']
    output:
        metrics_test='{output_dir}/summary/{cross_validation}/metrics.test.txt',
        metrics_train='{output_dir}/summary/{cross_validation}/metrics.train.txt'
    script:
        'scripts/summarize_cross_validation.py'