import ssl import requests as r from decimal import * import numpy as np def add_annotations(dataframe): print('Downloading UniProt sequence annotations...\n') ssl._create_default_https_context = ssl._create_unverified_context original_annot_name = ['DISULFID', 'INIT_MET', 'INTRAMEM', 'VARIANT', 'DNA_BIND', 'ACT_SITE', 'NP_BIND', 'LIPID', 'SITE', 'TRANSMEM', 'CROSSLNK', 'MUTAGEN', 'STRAND', 'HELIX', 'TURN', 'METAL', 'REPEAT', 'TOPO_DOM', 'CA_BIND', 'BINDING', 'REGION', 'SIGNAL', 'MOD_RES', 'ZN_FING', 'MOTIF', 'COILED', 'PEPTIDE', 'TRANSIT', 'CARBOHYD', 'PROPEP'] annotation_list = ['disulfide', 'intMet', 'intramembrane', 'naturalVariant', 'dnaBinding', 'activeSite', 'nucleotideBinding', 'lipidation', 'site', 'transmembrane', 'crosslink', 'mutagenesis', 'strand', 'helix', 'turn', 'metalBinding', 'repeat', 'topologicalDomain', 'caBinding', 'bindingSite', 'region', 'signalPeptide', 'modifiedResidue', 'zincFinger', 'motif', 'coiledCoil', 'peptide', 'transitPeptide', 'glycosylation', 'propeptide'] dataframe = dataframe.reset_index().drop(['index'], axis=1) for annot in original_annot_name: dataframe[annot] = '' for protein in list(set(dataframe.uniprotID.to_list())): print('Downloading annotations for ' + protein) uniprot_entry = r.get("http://www.uniprot.org/uniprot/" + protein + ".txt") uniprot_entry = uniprot_entry.text.split('\n') annot_for_protein = [] for annotation in original_annot_name: for line in uniprot_entry: if annotation.strip() in line and line.startswith( 'FT') and 'evidence' not in line and 'ECO' not in line and 'note' not in line: annot_for_protein.append(list(filter(None, line.split(' ')))[1:]) for select in annot_for_protein: if select[0] not in dataframe.columns: dataframe.loc[dataframe.uniprotID == protein, select[0]] = str((select[1] + '; ')) else: dataframe.loc[dataframe.uniprotID == protein, select[0]] += str((select[1] + '; ')) for i in range(len(original_annot_name)): dataframe = dataframe.rename(columns={original_annot_name[i]: annotation_list[i]}) # Fix annotation positions print('Processing positions...\n') for i in dataframe.index: for annot in dataframe.columns[-30:]: if annot != 'disulfide': if dataframe.at[i, annot] != 'nan': dataframe.at[i, annot] = ([x for x in [k.strip() for k in dataframe.at[i, annot].split(';')] if x]) if '..' not in str(dataframe.at[i, annot]): pass elif '..' in str(dataframe.at[i, annot]): dataframe.at[i, annot] = str(dataframe.at[i, annot]).replace('..', '-') else: disulfide_annot = [] if dataframe.at[i, annot] != 'nan': dataframe.at[i, annot]= dataframe.at[i, annot].split(';') dataframe.at[i, annot] = [i.split('..') for i in dataframe.at[i, annot]] dataframe.at[i, annot] =[e for v in dataframe.at[i, annot] for e in v] dataframe.at[i, annot] = [i for i in dataframe.at[i, annot] if i != ' '] # Add binary annotations print('Adding binary annotations...\n') dataframe = dataframe.astype('str') for i in dataframe.index: for k in annotation_list: # get the positions of each attribute as a list txt = k + 'Binary' dataframe.at[i, txt] = Decimal('nan') try: for positions in dataframe.at[i, k].split(','): position = positions.strip('[').strip(']').replace("'", "") if position != 'nan' and position != '' and '-' not in position and int( dataframe.at[i, 'pos']) == int(position): dataframe.at[i, txt] = '1' break elif position != 'nan' and position != '' and '-' not in position and int( dataframe.at[i, 'pos']) != int(position): dataframe.at[i, txt] = '0' elif position != 'nan' and position != '' and '-' in position: if int(position.split('-')[0]) < int(dataframe.at[i, 'pos']) < int(position.split('-')[1]): dataframe.at[i, txt] = '1' break else: dataframe.at[i, txt] = '0' except: ValueError # Final corrections dataframe = dataframe.replace({'[\'?\']': 'nan'}) dataframe = dataframe.replace({'[]': 'nan'}) return dataframe