callahantiff · January 29, 2022 20:48 · callahantiff · Jan 29, 2022
diff --git a/PheKnowLator_PE_Evaluation.py b/PheKnowLator_PE_Evaluation.py
 # import needed libraries
 import matplotlib.pyplot as plt
 import matplotlib
 matplotlib.style.use('ggplot')
 import matplotlib.patches as mpatches
 import numpy as np
 import pandas as pd
 import pickle
 import random
 from sklearn.manifold import TSNE
 from sklearn.decomposition import PCA
 from scipy.sparse import csr_matrix
 from sklearn.metrics.pairwise import cosine_similarity
 from sklearn.decomposition import TruncatedSVD
 from tqdm import tqdm


 def similarity_search(matrix, index_pat, top_n):
    """
    Function takes as input a tfidf matrix, an integer representing a patient id, and an integer representing the
    number of similar patients to return. The function uses this information and calculates the cosine similarity
    between the index patient and all other included patients. The results are sorted and returned as a list of
    lists where each list contains a patient identifier and the cosine similarity score the top set of similar as
    indicated by the input argument are returned.
    :param tfidf_matrix: where each row represents a patient and each column represents a concept and counts are
    weighted by TF-IDF
    :param index_patient: an integer representing a patient id
    :param top_n: an integer representing the number of similar patients to return
    :return:
        similar_patients: a list of lists where each list contains a patient identifier and the cosine similarity
        score the top set of similar as indicated by the input argument are returned
    """
    # http://markhneedham.com/blog/2016/07/27/scitkit-learn-tfidf-and-cosine-similarity-for-computer-science-papers/

    # calculate similarity
    cosine_similarities = cosine_similarity(matrix[index_pat:index_pat + 1], matrix).flatten()
    rel_pat_indices = [i for i in cosine_similarities.argsort()[::-1] if i != index_pat]
    similar_patients = [(patient, cosine_similarities[patient]) for patient in rel_pat_indices][0:top_n]

    return similar_patients
  
  def main():
    
    #######################################################################################
    # TASK 1 - Show that you can better understand ignorome genes using the pkt embeddings
    #######################################################################################
    # read in kg embeddings
    kg_file_dir = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/knowledge/'
    kg = pd.read_pickle(kg_file_dir + 'ALL_KG_res_df')
    kg_labels = pd.read_csv(kg_file_dir + 'KG_Node_Metadata.csv', header=0)
    kg = kg.merge(kg_labels[['id', 'label']], left_on='id', right_on='id', how='left').reset_index(drop=True)

    # from rdflib.namespace import OWL, RDF, RDFS  # type: ignore
    # from rdflib import BNode, Graph, Literal, Namespace, URIRef  # type: ignore
    # graph = Graph().parse(kg_file_dir + 'label_files/go_with_imports.owl')
    # go_typing = {
    #     x: str(list(graph.objects(x, URIRef('http://www.geneontology.org/formats/oboInOwl#hasOBONamespace')))[0])
    #     for x in set(x for x in graph.subjects(RDF.type, OWL.Class)) if str(x).split('/')[-1].startswith('GO')}

    # create subsets
    kg_no_genes = kg[kg['grp'] != 'Genes'].reset_index(drop=True)
    kg_no_genes['grp2'] = ['Entity'] * len(kg_no_genes)
    kg_genes = kg[kg['grp'] == 'Genes'].reset_index(drop=True)
    kg_genes['grp2'] = ['Genes'] * len(kg_genes)

    # ignorome genes
    file_dir = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/processed_results/PE_GeneExpression/'
    ignorome_genes = pd.read_csv(file_dir + 'pe_ignorome.csv', sep=',', header=0)
    ig_genes = ignorome_genes[ignorome_genes['IGNOROME'].notna()]['IGNOROME'].astype(int).astype(str).to_frame()
    kw_genes = ignorome_genes[ignorome_genes['KNOW'].notna()]['KNOW'].astype(int).astype(str).to_frame()
    kw_genes = ignorome_genes[ignorome_genes['BOTH'].notna()]['BOTH'].astype(int).astype(str).to_frame()
    updt_groups = ['PE' if kg.iloc[x]['id'] in set(ig_genes['IGNOROME']) else kg.iloc[x]['grp'] for x in range(0, len(kg))]

    # map embeddings to kg identifiers
    ig_gene_embeddings = ig_genes.merge(kg, left_on='IGNOROME', right_on='id', how='left').reset_index(drop=True)
    ig_gene_embeddings['grp2'] = ['Ignorome'] * len(ig_genes)
    kw_gene_embeddings = kw_genes.merge(kg, left_on='KNOW', right_on='id', how='left').reset_index(drop=True)
    kw_gene_embeddings['grp2'] = ['Knowome'] * len(kw_genes)

    # update labels for KG
    kg_no_genes = pd.concat([kg_no_genes, ig_gene_embeddings, kw_gene_embeddings])
    kg_no_tsne = 'ignorome_all_tsne.npy'
    kg_all_tsne = pd.concat([kg_genes, ig_gene_embeddings, kw_gene_embeddings])
    kg_all_genes = 'ignorome_pe_all_genes_tsne.npy'
    pe_genes = pd.concat([ig_gene_embeddings, kw_gene_embeddings])
    pe_gene_tsne = 'ignorome_pe_tsne.npy'

    # prep data for t-SNE
    matrix = [list(x) for x in kg_no_genes['embeds']]
    matrix = [list(x) for x in kg_all_genes['embeds']]
    matrix = [list(x) for x in pe_genes['embeds']]
    full_matrix = [list(x) for x in kg['embeds']]

    # create t-sne
    X_reduced = PCA(n_components=25, random_state=1).fit_transform(matrix)
    # X_reduced = TruncatedSVD(n_components=50, random_state=1, algorithm='arpack').fit_transform(matrix)
    X_embedded = TSNE(n_components=2, random_state=1, verbose=True, perplexity=50.0).fit_transform(X_reduced)
    np.save(file_dir + 'pca_pkt_embeddings_2d', X_embedded)
    X_embedded = np.load(file_dir + pe_gene_tsne)

    # set up colors and legend labels
    names = {'Edge': 'Edge', 'Gene Ontology': 'Gene Ontology', 'Drugs': 'Drugs', 'Diseases': 'Diseases', 'Pathways':
        'Pathways', 'Genes': 'Genes', 'Phenotypes': 'Phenotypes', 'PE': 'PE Ignorome'}
    colors = {'Diseases': 'paleturquoise', 'Drugs': 'lavenderblush', 'Gene Ontology': 'palegreen', 'Genes': 'lightgrey',
              'Pathways': 'mistyrose', 'Phenotypes': 'lavender', 'PE Ignorome': 'cornsilk'}
    dis = mpatches.Patch(color='deepskyblue', label='Disease'); drg = mpatches.Patch(color='pink', label='Drug')
    go = mpatches.Patch(color='lightgreen', label='GO'); ge = mpatches.Patch(color='dimgray', label='Gene')
    pat = mpatches.Patch(color='crimson', label='Pathway'); phe = mpatches.Patch(color='purple', label='Phenotype')
    pe = mpatches.Patch(color='goldenrod', label='Ignorome')
    # create data frame that has the result of the MDS plus the cluster numbers and titles
    df = pd.DataFrame(dict(x=X_embedded[:, 0], y=X_embedded[:, 1], group=list(updt_groups)))
    groups = df.groupby('group')
    fig, ax = plt.subplots(figsize=(13, 10))
    for x, grp in groups:
        if x == 'Gene Ontology':
            ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='forestgreen', alpha=0.6)
        if x == 'Genes':
            ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='dimgray', alpha=0.6)
        if x == 'Drugs':
            ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='hotpink', alpha=0.6)
        if x == 'Phenotypes':
            ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='purple', alpha=0.6)
        if x == 'Pathways':
            ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='crimson', alpha=0.6)
        if x == 'Diseases':
            ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='deepskyblue', alpha=0.6)
        if x == 'PE':
            ax.plot(grp.x, grp.y, marker='*', linestyle='', ms=12, label=names['PE'], color=colors['PE Ignorome'], mec='darkgoldenrod', alpha=0.6)
    plt.legend(handles=[dis, drg, go, ge, pat, phe, pe], fontsize=12, frameon=False, loc="lower center", ncol=7)
    ax.tick_params(labelsize=13)
    m0 = -55; m1 = 55; plt.ylim(m0, m1); plt.xlim(m0, m1)
    plt.savefig(file_dir + 'PKT_PEIgnorome_tSNE.png', bbox_inches='tight')
    plt.show(); plt.close()

    # create dicts to help with using matrix indices
    embedding_idx = kg['id'].to_dict(); embedding_id_grp = kg['grp'].to_dict()
    pe_gene_idx = {k: v for k, v in embedding_idx.items() if v in set(ig_genes['IGNOROME'])}
    kg_node_labels = {row['id']: row['label'] for idx, row in kg.iterrows() if row['grp'] != 'Edge'}

    # convert embeddings to compressed sparse matrix
    kg_matrix = csr_matrix([list(x) for x in kg['embeds']])

    # find similar entities
    similar_entities = {}
    for gene_idx, gene_id in tqdm(pe_gene_idx.items()):
        hit_dict = {}; gene_label = kg_node_labels[gene_id]
        matches = [x for x in similarity_search(kg_matrix, gene_idx, 200) if x[0] not in pe_gene_idx.keys()][0:100]
        if len(matches) < 100: break
        for x, score in matches:
            if embedding_id_grp[x] != 'Edge':
                match_id = embedding_idx[x]; node_grp = embedding_id_grp[x]; match_label = kg_node_labels[match_id]
                if node_grp in hit_dict.keys(): hit_dict[node_grp] += [[match_id, match_label, round(score, 3)]]
                else: hit_dict[node_grp] = [[match_id, match_label, round(score, 3)]]
        hit_dict['gene_label'] = gene_label
        similar_entities[gene_id] = hit_dict

    # k-means clustering -- unsupervised to find what the genes group with
    annotation_counts = {'Drugs': {}, 'Diseases': {}, 'Genes': {}, 'Pathways': {}, 'Phenotypes': {}, 'Gene Ontology': {}}
    master_list = dict()
    for k, v in tqdm(similar_entities.items()):
        gene_label = v['gene_label']; val = {x: y for x, y in v.items() if x != 'gene_label'}
        master_list[k] = {'Drugs': [], 'Diseases': [], 'Genes': [], 'Pathways': [], 'Phenotypes': [], 'Gene Ontology': []}
        for x in val.keys():
            for i in val[x]:
                if i[0] in annotation_counts[x].keys():
                    annotation_counts[x][i[0]]['scores'].append(i[-1])
                    annotation_counts[x][i[0]]['counts'] += 1
                    annotation_counts[x][i[0]]['genes'].append('{} ({})'.format(k, gene_label))
                else:
                    annotation_counts[x][i[0]] = {
                        'scores': [i[-1]], 'counts': 1, 'genes': ['{} ({})'.format(k, gene_label)]}
                master_list[k][x].append(i)

    # print top annotated entities
    for k, v in annotation_counts.items():
        for x in ['Drugs', 'Diseases', 'Genes', 'Gene Ontology', 'Pathways', 'Phenotypes']:
            k = x; v = annotation_counts[k]; overall_scores = [i for j in [v[x]['scores'] for x in v.keys()] for i in j]
            hs = sorted(list(set(overall_scores)), reverse=True)
            y = [((x, kg_node_labels[x]), v[x]['scores']) for x in v.keys()
                 if len([g for g in v[x]['scores'] if g in hs]) > 0]
            with open(file_dir + 'pkt_validation/pe_pkt_most_similar_{}.txt'.format(x), 'w') as f:
                for z in y:
                    hits = ', '.join(v[z[0][0]]['genes'])
                    try: f.write(str(z[0][0]) + '*' + str(z[0][1]) + '*' + str(z[1][0]) + '*' + hits + '\n')
                    except AttributeError: pass
            # printing most frequent by count
            count_list = [v[x]['counts'] for x in v.keys()]
            entity_counts = sorted(count_list, reverse=True); min_cnt = entity_counts[-1]; max_cnt = entity_counts[0]
            cut_off = [max_cnt] if entity_counts.count(max_cnt) > 1 else entity_counts
            he = [((x, kg_node_labels[x]), v[x]['counts']) for x in v.keys() if v[x]['counts'] in entity_counts]
            with open(file_dir + 'pkt_validation/pe_pkt_most_frequent_{}.txt'.format(x), 'w') as f:
                for u in he:
                    try: f.write(str(u[0][0]) + '*' + str(u[0][1]) + '*' + str(u[1]) + '\n')
                    except AttributeError: pass
 

    #######################################################################################
    # TASK 2 - Perfoming Modified Enrichment Analysis and Statistical Testing
    #######################################################################################
    
    write_loc = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/processed_results/pe_gene_testing/'
    gold_standard_annotations = {
        'Diseases': {'DOID_9870', 'DOID_1700', 'DOID_681', 'DOID_332', 'DOID_1094', 'DOID_5158', 'DOID_1926', 'DOID_9452'},
        'Drugs': {'D005486', 'C445526', 'C010634', 'D004962', 'C059714', 'D006003', 'C098288', 'D018927', 'D008277',
                  'D000068900'},
        'Genes': {'5352', '27094', '10576', '4709', '64983', '3047', '5931', '148867', '204962', '58480'},
        'Gene Ontology': {'GO_0070125', 'GO_0006882', 'GO_0005747', 'GO_0031462', 'GO_0005833', 'GO_0005516', 'GO_0007062',
                          'GO_0042393', 'GO_0000398', 'GO_0051015'},
        'Pathways': {'R-HSA-6799198', 'R-HSA-5419276', 'R-HSA-611105', 'R-HSA-1566948', 'R-HSA-3906995', 'R-HSA-194840',
                     'R-HSA-391251', 'R-HSA-2500257', 'R-HSA-913531', 'R-HSA-212165'},
        'Phenotypes': {'HP_0008316', 'HP_0002725', 'HP_0008344', 'HP_0011904', 'HP_0001935'}
    }
    pickle.dump(gold_standard_annotations, open(write_loc + 'Expert_Reviewed_Annotations', 'wb'))

    # read kg files
    # kg_file_dir = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/knowledge/'
    # kg = pd.read_pickle(kg_file_dir + 'ALL_KG_res_df')
    # kg_labels = pd.read_csv(kg_file_dir + 'KG_Node_Metadata.csv', header=0)
    # kg = kg.merge(kg_labels[['id', 'label']], left_on='id', right_on='id', how='left').reset_index(drop=True)
    # set-up gene lists
    # ig_genes = ignorome_genes[ignorome_genes['IGNOROME'].notna()]['IGNOROME'].astype(int).astype(str).to_frame()
    # kw_genes = ignorome_genes[ignorome_genes['KNOW'].notna()]['KNOW'].astype(int).astype(str).to_frame()
    # both_genes = ignorome_genes[ignorome_genes['BOTH'].notna()]['BOTH'].astype(int).astype(str).to_frame()
    # pe_gene_idx = set(ig_genes['IGNOROME']) | set(kw_genes['KNOW']) | set(both_genes['BOTH'])
    # create dicts to help with using matrix indices
    # embedding_idx = kg['id'].to_dict(); embedding_id_grp = kg['grp'].to_dict()
    # rev_embedding_idx = {v: k for k, v in embedding_idx.items()}
    # kg_node_labels = {row['id']: row['label'] for idx, row in kg.iterrows() if row['grp'] != 'Edge'}
    # lab_dicts = {'embedding_idx': embedding_idx, 'rev_embedding_idx': rev_embedding_idx, 'kg_node_labels' : kg_node_labels,
    #              'embedding_id_grp': embedding_id_grp}
    # pickle.dump(lab_dicts, open(write_loc + 'idx_dicts', 'wb'))
    lab_dict = pickle.load(open(write_loc + 'idx_dicts', 'rb'))
    embedding_idx = lab_dict['embedding_idx']; rev_embedding_idx = lab_dict['rev_embedding_idx']
    kg_node_labels = lab_dict['kg_node_labels']; embedding_id_grp = lab_dict['embedding_id_grp']

    # convert embeddings to compressed sparse matrix
    # kg_matrix = csr_matrix([list(x) for x in kg['embeds']])
    # pickle.dump(kg_matrix, open(write_loc + 'pkt_kg_matrix', 'wb'))
    kg_matrix = pickle.load(open(write_loc + 'pkt_kg_matrix', 'rb'))

    # pull 1,000 samples (with replacement) from all genes other than PE
    # sample_set = set(kg[kg['grp'] == 'Genes']['id']) - pe_gene_idx  # 22385
    # gene_samples = [random.choices(list(sample_set), k=445) for _ in range(0, 1000)]

    # gene_samp = set([i for u in gene_samples for i in u])
    # updated = set([x for x in gene_samp if x not in gene_samp])
    # pickle.dump(gene_samples, open(write_loc + 'sample_list', 'wb'))
    gene_samples = pickle.load(open(write_loc + 'sample_list', 'rb'))
    sample_start = 0; sample_end = 101; genes = gene_samples[sample_start:sample_end]
    # sample_start = 100; sample_end = 201; genes = gene_samples[sample_start:sample_end]

    # find similar entities
    idx = list(range(0, 101))
    for samp in tqdm(range(sample_start, sample_end)):
        sampled_results = {}
        for gene_id in tqdm(genes[idx.pop(0)]):
            print(gene_id, len(genes))
            gene_idx = rev_embedding_idx[gene_id]; hit_dict = {}; gene_label = kg_node_labels[gene_id]
            matches = similarity_search(kg_matrix, gene_idx, 100); sampled_results[samp] = {}
            for x, score in matches:
                if embedding_id_grp[x] != 'Edge':
                    match_id = embedding_idx[x]; node_grp = embedding_id_grp[x]; match_label = kg_node_labels[match_id]
                    if node_grp in hit_dict.keys(): hit_dict[node_grp] += [[match_id, match_label, round(score, 3)]]
                    else: hit_dict[node_grp] = [[match_id, match_label, round(score, 3)]]
            hit_dict['gene_label'] = gene_label
            sampled_results[samp][gene_id] = hit_dict

    # reads in the annotation runs and organizes them by domain --> hits --> entity-scores
    from glob import glob
    f = glob(write_loc + 'sample_sets/*sample*')[:-1]
    sampled_results = {k: v for d in [pickle.load(open(x, 'rb')) for x in tqdm(f)] for k, v in d.items()}
    annotation_counts = {'Drugs': {}, 'Diseases': {}, 'Genes': {}, 'Pathways': {}, 'Phenotypes': {}, 'Gene Ontology': {}}
    for s in tqdm(list(sampled_results.keys())[0:1000]):
        annotation_counts['Drugs'][s] = []; annotation_counts['Diseases'][s] = []
        annotation_counts['Genes'][s] = []; annotation_counts['Gene Ontology'][s] = []
        annotation_counts['Pathways'][s] = []; annotation_counts['Phenotypes'][s] = []
        for k, v in sampled_results[s].items():
            for x in v.keys():
                if x != 'gene_label':
                    for i in v[x]:
                        annotation_counts[x][s] += ['{}:{}'.format(i[0], i[2])]

    # for any repeating annotations within the same run, this method condenses them by the max similarity score
    hits1 = {'Drugs': [], 'Diseases': [], 'Genes': [], 'Pathways': [], 'Phenotypes': [], 'Gene Ontology': []}
    for k, v in tqdm(annotation_counts.items()):
        hits1[k] = {}; ent_dict = {}
        for run, out in v.items():
            for i in out:
                term, score = i.split(':')
                if term in ent_dict.keys(): ent_dict[term] = max([ent_dict[term], float(score)])
                else: ent_dict[term] = float(score)
        hits1[k] = ['{}:{}'.format(k, v) for k, v in ent_dict.items()]

    # gets final hits and scores tests
    hits = {'Drugs': [], 'Diseases': [], 'Genes': [], 'Pathways': [], 'Phenotypes': [], 'Gene Ontology': []}
    for k, v in tqdm(hits1.items()):
        if k == 'Drugs': cut =118
        if k == 'Diseases': cut = 15
        if k == 'Genes': cut = 76
        if k == 'Pathways': cut = 44
        if k == 'Phenotypes': cut = 11
        if k == 'Gene Ontology': cut = 62
        gs = gold_standard_annotations[k]; entity_list, entity_scores = [], []
        tup_list = [tuple(x.split(':')) for x in v]
        sorted_tuples = sorted(tup_list, key=lambda x: x[1], reverse=True)
        hits_to_search = [x[0] for x in sorted_tuples]; hit_scores = [float(x[1]) for x in sorted_tuples]
        # hit = list(gs.intersection(set(hits_to_search[0:len(gs)])))
        hit = list(gs.intersection(set(hits_to_search[0:cut])))
        if len(hit) > 0: hits[k] += ['{}: {}'.format(x, kg_node_labels[x].lower()) for x in hit]
        # entity_list += hits_to_search
        # entity_scores += hit_scores
        #
        # print('DOMAIN: {}'.format(k))
        # print('Entites: {} Total Entities; {} Unique'.format(len(entity_list), len(set(entity_list))))
        # print('Similarity Score: {} Min; {} Max; {} Average'.format(round(min(entity_scores), 3),
        #                                                             round(max(entity_scores), 3),
        #                                                             round(sum(entity_scores)/len(entity_scores), 3)))
    for k, v in hits.items():
        print('{}: {}, p-value={}'.format(k.upper(), len(v), len(v)/float(1000)))
	# import needed libraries
	import matplotlib.pyplot as plt
	import matplotlib
	matplotlib.style.use('ggplot')
	import matplotlib.patches as mpatches
	import numpy as np
	import pandas as pd
	import pickle
	import random
	from sklearn.manifold import TSNE
	from sklearn.decomposition import PCA
	from scipy.sparse import csr_matrix
	from sklearn.metrics.pairwise import cosine_similarity
	from sklearn.decomposition import TruncatedSVD
	from tqdm import tqdm


	def similarity_search(matrix, index_pat, top_n):
	"""
	Function takes as input a tfidf matrix, an integer representing a patient id, and an integer representing the
	number of similar patients to return. The function uses this information and calculates the cosine similarity
	between the index patient and all other included patients. The results are sorted and returned as a list of
	lists where each list contains a patient identifier and the cosine similarity score the top set of similar as
	indicated by the input argument are returned.
	:param tfidf_matrix: where each row represents a patient and each column represents a concept and counts are
	weighted by TF-IDF
	:param index_patient: an integer representing a patient id
	:param top_n: an integer representing the number of similar patients to return
	:return:
	similar_patients: a list of lists where each list contains a patient identifier and the cosine similarity
	score the top set of similar as indicated by the input argument are returned
	"""
	# http://markhneedham.com/blog/2016/07/27/scitkit-learn-tfidf-and-cosine-similarity-for-computer-science-papers/

	# calculate similarity
	cosine_similarities = cosine_similarity(matrix[index_pat:index_pat + 1], matrix).flatten()
	rel_pat_indices = [i for i in cosine_similarities.argsort()[::-1] if i != index_pat]
	similar_patients = [(patient, cosine_similarities[patient]) for patient in rel_pat_indices][0:top_n]

	return similar_patients

	def main():

	#######################################################################################
	# TASK 1 - Show that you can better understand ignorome genes using the pkt embeddings
	#######################################################################################
	# read in kg embeddings
	kg_file_dir = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/knowledge/'
	kg = pd.read_pickle(kg_file_dir + 'ALL_KG_res_df')
	kg_labels = pd.read_csv(kg_file_dir + 'KG_Node_Metadata.csv', header=0)
	kg = kg.merge(kg_labels[['id', 'label']], left_on='id', right_on='id', how='left').reset_index(drop=True)

	# from rdflib.namespace import OWL, RDF, RDFS # type: ignore
	# from rdflib import BNode, Graph, Literal, Namespace, URIRef # type: ignore
	# graph = Graph().parse(kg_file_dir + 'label_files/go_with_imports.owl')
	# go_typing = {
	# x: str(list(graph.objects(x, URIRef('http://www.geneontology.org/formats/oboInOwl#hasOBONamespace')))[0])
	# for x in set(x for x in graph.subjects(RDF.type, OWL.Class)) if str(x).split('/')[-1].startswith('GO')}

	# create subsets
	kg_no_genes = kg[kg['grp'] != 'Genes'].reset_index(drop=True)
	kg_no_genes['grp2'] = ['Entity'] * len(kg_no_genes)
	kg_genes = kg[kg['grp'] == 'Genes'].reset_index(drop=True)
	kg_genes['grp2'] = ['Genes'] * len(kg_genes)

	# ignorome genes
	file_dir = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/processed_results/PE_GeneExpression/'
	ignorome_genes = pd.read_csv(file_dir + 'pe_ignorome.csv', sep=',', header=0)
	ig_genes = ignorome_genes[ignorome_genes['IGNOROME'].notna()]['IGNOROME'].astype(int).astype(str).to_frame()
	kw_genes = ignorome_genes[ignorome_genes['KNOW'].notna()]['KNOW'].astype(int).astype(str).to_frame()
	kw_genes = ignorome_genes[ignorome_genes['BOTH'].notna()]['BOTH'].astype(int).astype(str).to_frame()
	updt_groups = ['PE' if kg.iloc[x]['id'] in set(ig_genes['IGNOROME']) else kg.iloc[x]['grp'] for x in range(0, len(kg))]

	# map embeddings to kg identifiers
	ig_gene_embeddings = ig_genes.merge(kg, left_on='IGNOROME', right_on='id', how='left').reset_index(drop=True)
	ig_gene_embeddings['grp2'] = ['Ignorome'] * len(ig_genes)
	kw_gene_embeddings = kw_genes.merge(kg, left_on='KNOW', right_on='id', how='left').reset_index(drop=True)
	kw_gene_embeddings['grp2'] = ['Knowome'] * len(kw_genes)

	# update labels for KG
	kg_no_genes = pd.concat([kg_no_genes, ig_gene_embeddings, kw_gene_embeddings])
	kg_no_tsne = 'ignorome_all_tsne.npy'
	kg_all_tsne = pd.concat([kg_genes, ig_gene_embeddings, kw_gene_embeddings])
	kg_all_genes = 'ignorome_pe_all_genes_tsne.npy'
	pe_genes = pd.concat([ig_gene_embeddings, kw_gene_embeddings])
	pe_gene_tsne = 'ignorome_pe_tsne.npy'

	# prep data for t-SNE
	matrix = [list(x) for x in kg_no_genes['embeds']]
	matrix = [list(x) for x in kg_all_genes['embeds']]
	matrix = [list(x) for x in pe_genes['embeds']]
	full_matrix = [list(x) for x in kg['embeds']]

	# create t-sne
	X_reduced = PCA(n_components=25, random_state=1).fit_transform(matrix)
	# X_reduced = TruncatedSVD(n_components=50, random_state=1, algorithm='arpack').fit_transform(matrix)
	X_embedded = TSNE(n_components=2, random_state=1, verbose=True, perplexity=50.0).fit_transform(X_reduced)
	np.save(file_dir + 'pca_pkt_embeddings_2d', X_embedded)
	X_embedded = np.load(file_dir + pe_gene_tsne)

	# set up colors and legend labels
	names = {'Edge': 'Edge', 'Gene Ontology': 'Gene Ontology', 'Drugs': 'Drugs', 'Diseases': 'Diseases', 'Pathways':
	'Pathways', 'Genes': 'Genes', 'Phenotypes': 'Phenotypes', 'PE': 'PE Ignorome'}
	colors = {'Diseases': 'paleturquoise', 'Drugs': 'lavenderblush', 'Gene Ontology': 'palegreen', 'Genes': 'lightgrey',
	'Pathways': 'mistyrose', 'Phenotypes': 'lavender', 'PE Ignorome': 'cornsilk'}
	dis = mpatches.Patch(color='deepskyblue', label='Disease'); drg = mpatches.Patch(color='pink', label='Drug')
	go = mpatches.Patch(color='lightgreen', label='GO'); ge = mpatches.Patch(color='dimgray', label='Gene')
	pat = mpatches.Patch(color='crimson', label='Pathway'); phe = mpatches.Patch(color='purple', label='Phenotype')
	pe = mpatches.Patch(color='goldenrod', label='Ignorome')
	# create data frame that has the result of the MDS plus the cluster numbers and titles
	df = pd.DataFrame(dict(x=X_embedded[:, 0], y=X_embedded[:, 1], group=list(updt_groups)))
	groups = df.groupby('group')
	fig, ax = plt.subplots(figsize=(13, 10))
	for x, grp in groups:
	if x == 'Gene Ontology':
	ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='forestgreen', alpha=0.6)
	if x == 'Genes':
	ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='dimgray', alpha=0.6)
	if x == 'Drugs':
	ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='hotpink', alpha=0.6)
	if x == 'Phenotypes':
	ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='purple', alpha=0.6)
	if x == 'Pathways':
	ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='crimson', alpha=0.6)
	if x == 'Diseases':
	ax.plot(grp.x, grp.y, marker='o', linestyle='', ms=4, label=names[x], color=colors[x], mec='deepskyblue', alpha=0.6)
	if x == 'PE':
	ax.plot(grp.x, grp.y, marker='*', linestyle='', ms=12, label=names['PE'], color=colors['PE Ignorome'], mec='darkgoldenrod', alpha=0.6)
	plt.legend(handles=[dis, drg, go, ge, pat, phe, pe], fontsize=12, frameon=False, loc="lower center", ncol=7)
	ax.tick_params(labelsize=13)
	m0 = -55; m1 = 55; plt.ylim(m0, m1); plt.xlim(m0, m1)
	plt.savefig(file_dir + 'PKT_PEIgnorome_tSNE.png', bbox_inches='tight')
	plt.show(); plt.close()

	# create dicts to help with using matrix indices
	embedding_idx = kg['id'].to_dict(); embedding_id_grp = kg['grp'].to_dict()
	pe_gene_idx = {k: v for k, v in embedding_idx.items() if v in set(ig_genes['IGNOROME'])}
	kg_node_labels = {row['id']: row['label'] for idx, row in kg.iterrows() if row['grp'] != 'Edge'}

	# convert embeddings to compressed sparse matrix
	kg_matrix = csr_matrix([list(x) for x in kg['embeds']])

	# find similar entities
	similar_entities = {}
	for gene_idx, gene_id in tqdm(pe_gene_idx.items()):
	hit_dict = {}; gene_label = kg_node_labels[gene_id]
	matches = [x for x in similarity_search(kg_matrix, gene_idx, 200) if x[0] not in pe_gene_idx.keys()][0:100]
	if len(matches) < 100: break
	for x, score in matches:
	if embedding_id_grp[x] != 'Edge':
	match_id = embedding_idx[x]; node_grp = embedding_id_grp[x]; match_label = kg_node_labels[match_id]
	if node_grp in hit_dict.keys(): hit_dict[node_grp] += [[match_id, match_label, round(score, 3)]]
	else: hit_dict[node_grp] = [[match_id, match_label, round(score, 3)]]
	hit_dict['gene_label'] = gene_label
	similar_entities[gene_id] = hit_dict

	# k-means clustering -- unsupervised to find what the genes group with
	annotation_counts = {'Drugs': {}, 'Diseases': {}, 'Genes': {}, 'Pathways': {}, 'Phenotypes': {}, 'Gene Ontology': {}}
	master_list = dict()
	for k, v in tqdm(similar_entities.items()):
	gene_label = v['gene_label']; val = {x: y for x, y in v.items() if x != 'gene_label'}
	master_list[k] = {'Drugs': [], 'Diseases': [], 'Genes': [], 'Pathways': [], 'Phenotypes': [], 'Gene Ontology': []}
	for x in val.keys():
	for i in val[x]:
	if i[0] in annotation_counts[x].keys():
	annotation_counts[x][i[0]]['scores'].append(i[-1])
	annotation_counts[x][i[0]]['counts'] += 1
	annotation_counts[x][i[0]]['genes'].append('{} ({})'.format(k, gene_label))
	else:
	annotation_counts[x][i[0]] = {
	'scores': [i[-1]], 'counts': 1, 'genes': ['{} ({})'.format(k, gene_label)]}
	master_list[k][x].append(i)

	# print top annotated entities
	for k, v in annotation_counts.items():
	for x in ['Drugs', 'Diseases', 'Genes', 'Gene Ontology', 'Pathways', 'Phenotypes']:
	k = x; v = annotation_counts[k]; overall_scores = [i for j in [v[x]['scores'] for x in v.keys()] for i in j]
	hs = sorted(list(set(overall_scores)), reverse=True)
	y = [((x, kg_node_labels[x]), v[x]['scores']) for x in v.keys()
	if len([g for g in v[x]['scores'] if g in hs]) > 0]
	with open(file_dir + 'pkt_validation/pe_pkt_most_similar_{}.txt'.format(x), 'w') as f:
	for z in y:
	hits = ', '.join(v[z[0][0]]['genes'])
	try: f.write(str(z[0][0]) + '' + str(z[0][1]) + '' + str(z[1][0]) + '*' + hits + '\n')
	except AttributeError: pass
	# printing most frequent by count
	count_list = [v[x]['counts'] for x in v.keys()]
	entity_counts = sorted(count_list, reverse=True); min_cnt = entity_counts[-1]; max_cnt = entity_counts[0]
	cut_off = [max_cnt] if entity_counts.count(max_cnt) > 1 else entity_counts
	he = [((x, kg_node_labels[x]), v[x]['counts']) for x in v.keys() if v[x]['counts'] in entity_counts]
	with open(file_dir + 'pkt_validation/pe_pkt_most_frequent_{}.txt'.format(x), 'w') as f:
	for u in he:
	try: f.write(str(u[0][0]) + '' + str(u[0][1]) + '' + str(u[1]) + '\n')
	except AttributeError: pass


	#######################################################################################
	# TASK 2 - Perfoming Modified Enrichment Analysis and Statistical Testing
	#######################################################################################

	write_loc = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/processed_results/pe_gene_testing/'
	gold_standard_annotations = {
	'Diseases': {'DOID_9870', 'DOID_1700', 'DOID_681', 'DOID_332', 'DOID_1094', 'DOID_5158', 'DOID_1926', 'DOID_9452'},
	'Drugs': {'D005486', 'C445526', 'C010634', 'D004962', 'C059714', 'D006003', 'C098288', 'D018927', 'D008277',
	'D000068900'},
	'Genes': {'5352', '27094', '10576', '4709', '64983', '3047', '5931', '148867', '204962', '58480'},
	'Gene Ontology': {'GO_0070125', 'GO_0006882', 'GO_0005747', 'GO_0031462', 'GO_0005833', 'GO_0005516', 'GO_0007062',
	'GO_0042393', 'GO_0000398', 'GO_0051015'},
	'Pathways': {'R-HSA-6799198', 'R-HSA-5419276', 'R-HSA-611105', 'R-HSA-1566948', 'R-HSA-3906995', 'R-HSA-194840',
	'R-HSA-391251', 'R-HSA-2500257', 'R-HSA-913531', 'R-HSA-212165'},
	'Phenotypes': {'HP_0008316', 'HP_0002725', 'HP_0008344', 'HP_0011904', 'HP_0001935'}
	}
	pickle.dump(gold_standard_annotations, open(write_loc + 'Expert_Reviewed_Annotations', 'wb'))

	# read kg files
	# kg_file_dir = 'resources/releases/v1.0.0/Dissertation/PheKnowLator Evaluation/knowledge/'
	# kg = pd.read_pickle(kg_file_dir + 'ALL_KG_res_df')
	# kg_labels = pd.read_csv(kg_file_dir + 'KG_Node_Metadata.csv', header=0)
	# kg = kg.merge(kg_labels[['id', 'label']], left_on='id', right_on='id', how='left').reset_index(drop=True)
	# set-up gene lists
	# ig_genes = ignorome_genes[ignorome_genes['IGNOROME'].notna()]['IGNOROME'].astype(int).astype(str).to_frame()
	# kw_genes = ignorome_genes[ignorome_genes['KNOW'].notna()]['KNOW'].astype(int).astype(str).to_frame()
	# both_genes = ignorome_genes[ignorome_genes['BOTH'].notna()]['BOTH'].astype(int).astype(str).to_frame()
	# pe_gene_idx = set(ig_genes['IGNOROME']) \| set(kw_genes['KNOW']) \| set(both_genes['BOTH'])
	# create dicts to help with using matrix indices
	# embedding_idx = kg['id'].to_dict(); embedding_id_grp = kg['grp'].to_dict()
	# rev_embedding_idx = {v: k for k, v in embedding_idx.items()}
	# kg_node_labels = {row['id']: row['label'] for idx, row in kg.iterrows() if row['grp'] != 'Edge'}
	# lab_dicts = {'embedding_idx': embedding_idx, 'rev_embedding_idx': rev_embedding_idx, 'kg_node_labels' : kg_node_labels,
	# 'embedding_id_grp': embedding_id_grp}
	# pickle.dump(lab_dicts, open(write_loc + 'idx_dicts', 'wb'))
	lab_dict = pickle.load(open(write_loc + 'idx_dicts', 'rb'))
	embedding_idx = lab_dict['embedding_idx']; rev_embedding_idx = lab_dict['rev_embedding_idx']
	kg_node_labels = lab_dict['kg_node_labels']; embedding_id_grp = lab_dict['embedding_id_grp']

	# convert embeddings to compressed sparse matrix
	# kg_matrix = csr_matrix([list(x) for x in kg['embeds']])
	# pickle.dump(kg_matrix, open(write_loc + 'pkt_kg_matrix', 'wb'))
	kg_matrix = pickle.load(open(write_loc + 'pkt_kg_matrix', 'rb'))

	# pull 1,000 samples (with replacement) from all genes other than PE
	# sample_set = set(kg[kg['grp'] == 'Genes']['id']) - pe_gene_idx # 22385
	# gene_samples = [random.choices(list(sample_set), k=445) for _ in range(0, 1000)]

	# gene_samp = set([i for u in gene_samples for i in u])
	# updated = set([x for x in gene_samp if x not in gene_samp])
	# pickle.dump(gene_samples, open(write_loc + 'sample_list', 'wb'))
	gene_samples = pickle.load(open(write_loc + 'sample_list', 'rb'))
	sample_start = 0; sample_end = 101; genes = gene_samples[sample_start:sample_end]
	# sample_start = 100; sample_end = 201; genes = gene_samples[sample_start:sample_end]

	# find similar entities
	idx = list(range(0, 101))
	for samp in tqdm(range(sample_start, sample_end)):
	sampled_results = {}
	for gene_id in tqdm(genes[idx.pop(0)]):
	print(gene_id, len(genes))
	gene_idx = rev_embedding_idx[gene_id]; hit_dict = {}; gene_label = kg_node_labels[gene_id]
	matches = similarity_search(kg_matrix, gene_idx, 100); sampled_results[samp] = {}
	for x, score in matches:
	if embedding_id_grp[x] != 'Edge':
	match_id = embedding_idx[x]; node_grp = embedding_id_grp[x]; match_label = kg_node_labels[match_id]
	if node_grp in hit_dict.keys(): hit_dict[node_grp] += [[match_id, match_label, round(score, 3)]]
	else: hit_dict[node_grp] = [[match_id, match_label, round(score, 3)]]
	hit_dict['gene_label'] = gene_label
	sampled_results[samp][gene_id] = hit_dict

	# reads in the annotation runs and organizes them by domain --> hits --> entity-scores
	from glob import glob
	f = glob(write_loc + 'sample_sets/sample')[:-1]
	sampled_results = {k: v for d in [pickle.load(open(x, 'rb')) for x in tqdm(f)] for k, v in d.items()}
	annotation_counts = {'Drugs': {}, 'Diseases': {}, 'Genes': {}, 'Pathways': {}, 'Phenotypes': {}, 'Gene Ontology': {}}
	for s in tqdm(list(sampled_results.keys())[0:1000]):
	annotation_counts['Drugs'][s] = []; annotation_counts['Diseases'][s] = []
	annotation_counts['Genes'][s] = []; annotation_counts['Gene Ontology'][s] = []
	annotation_counts['Pathways'][s] = []; annotation_counts['Phenotypes'][s] = []
	for k, v in sampled_results[s].items():
	for x in v.keys():
	if x != 'gene_label':
	for i in v[x]:
	annotation_counts[x][s] += ['{}:{}'.format(i[0], i[2])]

	# for any repeating annotations within the same run, this method condenses them by the max similarity score
	hits1 = {'Drugs': [], 'Diseases': [], 'Genes': [], 'Pathways': [], 'Phenotypes': [], 'Gene Ontology': []}
	for k, v in tqdm(annotation_counts.items()):
	hits1[k] = {}; ent_dict = {}
	for run, out in v.items():
	for i in out:
	term, score = i.split(':')
	if term in ent_dict.keys(): ent_dict[term] = max([ent_dict[term], float(score)])
	else: ent_dict[term] = float(score)
	hits1[k] = ['{}:{}'.format(k, v) for k, v in ent_dict.items()]

	# gets final hits and scores tests
	hits = {'Drugs': [], 'Diseases': [], 'Genes': [], 'Pathways': [], 'Phenotypes': [], 'Gene Ontology': []}
	for k, v in tqdm(hits1.items()):
	if k == 'Drugs': cut =118
	if k == 'Diseases': cut = 15
	if k == 'Genes': cut = 76
	if k == 'Pathways': cut = 44
	if k == 'Phenotypes': cut = 11
	if k == 'Gene Ontology': cut = 62
	gs = gold_standard_annotations[k]; entity_list, entity_scores = [], []
	tup_list = [tuple(x.split(':')) for x in v]
	sorted_tuples = sorted(tup_list, key=lambda x: x[1], reverse=True)
	hits_to_search = [x[0] for x in sorted_tuples]; hit_scores = [float(x[1]) for x in sorted_tuples]
	# hit = list(gs.intersection(set(hits_to_search[0:len(gs)])))
	hit = list(gs.intersection(set(hits_to_search[0:cut])))
	if len(hit) > 0: hits[k] += ['{}: {}'.format(x, kg_node_labels[x].lower()) for x in hit]
	# entity_list += hits_to_search
	# entity_scores += hit_scores
	#
	# print('DOMAIN: {}'.format(k))
	# print('Entites: {} Total Entities; {} Unique'.format(len(entity_list), len(set(entity_list))))
	# print('Similarity Score: {} Min; {} Max; {} Average'.format(round(min(entity_scores), 3),
	# round(max(entity_scores), 3),
	# round(sum(entity_scores)/len(entity_scores), 3)))
	for k, v in hits.items():
	print('{}: {}, p-value={}'.format(k.upper(), len(v), len(v)/float(1000)))