def test_create_dict_matrices_hb(systems, hb_twochains, hb_references): # Get the reference matrices ref_matrices = hb_references["matrices"]["two_chains"] # Get the dictionary containing the computed tables dict_tables = \ li.create_dict_tables(hb_twochains["table"]) # Get the dictionary containing the computed matrices dict_matrices = \ li.create_dict_matrices(hb_twochains["matrix"], dict_tables, systems["two_chains"]["pdb"]) # Dictionary mapping the groups of contacts (all, intrachain, # interchain) to the corresponding keys in the dictionary of # reference matrices groups2keys = { "all": "all", "A": "intra_a", "B": "intra_b", ("A", "B"): "inter_ab" } # Check the computed matrices against the references dict_matrices_test(groups2keys, dict_matrices, ref_matrices)
def main(): ######################### ARGUMENT PARSER ######################### description = "Interaction calculator" parser = argparse.ArgumentParser(description=description) #---------------------------- Top/traj ---------------------------# s_helpstr = "Topology file" parser.add_argument("-s", "--top", action="store", type=str, dest="top", default=None, help=s_helpstr) t_helpstr = "Trajectory file" parser.add_argument("-t", "--trj", action="store", type=str, dest="trj", default=None, help=t_helpstr) r_helpstr = "Reference structure" parser.add_argument("-r", "--ref", action="store", type=str, dest="ref", default=None, help=r_helpstr) #---------------------------- Analyses ---------------------------# m_helpstr = "Analyze side-chain centers-of-mass contacts (cmPSN)" parser.add_argument("-m", "--cmpsn", action="store_true", dest="do_cmpsn", help=m_helpstr) a_helpstr = "Analyze atomic contacts (acPSN)" parser.add_argument("-a", "--acpsn", action="store_true", dest="do_acpsn", help=a_helpstr) f_helpstr = "Analyze hydrophobic clusters" parser.add_argument("-f", "--hydrophobic", action="store_true", dest="do_hc", help=f_helpstr) b_helpstr = "Analyze salt bridges" parser.add_argument("-b", "--salt-bridges", action="store_true", dest="do_sb", help=b_helpstr) y_helpstr = "Analyze hydrogen bonds" parser.add_argument("-y", "--hydrogen-bonds", action="store_true", dest="do_hb", help=y_helpstr) p_helpstr = \ "Analyze interactions using the knowledge-based potential" parser.add_argument("-p", "--potential", action="store_true", dest="do_kbp", help=p_helpstr) #----------------------------- cmPSN -----------------------------# cmpsnco_default = 5.0 cmpsnco_helpstr = \ f"Distance cut-off for the cmPSN (default: {cmpsnco_default})" parser.add_argument("--cmpsn-co", "--cmpsn-cutoff", action="store", dest="cmpsn_co", type=float, default=cmpsnco_default, help=cmpsnco_helpstr) cmpsnperco_default = 0.0 cmpsnperco_helpstr = \ f"Minimum persistence for the cmPSN (default: " \ f"{cmpsnperco_default})" parser.add_argument("--cmpsn-perco", "--cmpsn-persistence-cutoff", action="store", type=float, dest="cmpsn_perco", default=cmpsnperco_default, help=cmpsnperco_helpstr) cmpsn_reslist = \ ["ALA", "CYS", "ASP", "GLU", "PHE", "HIS", "ILE", "LYS", "LEU", "MET", "ASN", "PRO", "GLN", "ARG", "SER", "THR", "VAL", "TRP", "TYR"] cmpsnres_helpstr = \ f"Comma-separated list of residues to be used when " \ f"calculating the cmPSN (default: {', '.join(cmpsn_reslist)})" parser.add_argument("--cmpsn-residues", action="store", type=str, dest="cmpsn_reslist", default=cmpsn_reslist, help=cmpsnres_helpstr) cmpsn_correction_choices = ["null", "rg"] cmpsn_correction_default = "null" cmpsn_correction_helpstr = \ f"Correction to be applied to the cmPSN (default: " \ f"{cmpsn_correction_default}" parser.add_argument("--cmpsn-correction", action="store", dest="cmpsn_correction", type=str, choices=cmpsn_correction_choices, default=cmpsn_correction_default, help=cmpsn_correction_helpstr) cmpsncsv_default = "cmpsn.csv" cmpsncsv_helpstr = \ f"Name of the CSV file where to store the list of contacts " \ f"found in the cmPSN (default: {cmpsncsv_default})" parser.add_argument("--cmpsn-csv", action="store", type=str, dest="cmpsn_csv", default=cmpsncsv_default, help=cmpsncsv_helpstr) cmpsngraph_helpstr = \ "File where to store the adjacency matrix for the " \ "interaction graph (cmPSN)" parser.add_argument("--cmpsn-graph", action="store", dest="cmpsn_graph", type=str, default=None, help=cmpsngraph_helpstr) #----------------------------- acPSN -----------------------------# acpsnco_default = 4.5 acpsnco_helpstr = \ f"Distance cut-off for the acPSN (default: {acpsnco_default})" parser.add_argument("--acpsn-co", "--acpsn-cutoff", action="store", dest="acpsn_co", type=float, default=acpsnco_default, help=acpsnco_helpstr) acpsnperco_default = 0.0 acpsnperco_helpstr = \ f"Minimum persistence for the acPSN (default: " \ f"{acpsnperco_default})" parser.add_argument("--acpsn-perco", "--acpsn-persistence-cutoff", action="store", type=float, dest="acpsn_perco", default=acpsnperco_default, help=acpsnperco_helpstr) acpsnproxco_default = 1 acpsnproxco_helpstr = \ f"Minimum sequence distance for the acPSN (default: " \ f"{acpsnproxco_default})" parser.add_argument("--acpsn-proxco", "--acpsn-sequence-cutoff", action="store", type=int, dest="acpsn_proxco", default=acpsnproxco_default, help=acpsnproxco_helpstr) acpsnimin_default = 3.0 acpsnimin_helpstr = \ f"Minimum interaction strength value for the acPSN " \ f"(default: {acpsnimin_default})" parser.add_argument("--acpsn-imin", "--acpsn-imin-cutoff", action="store", type=float, dest="acpsn_imin", default=acpsnimin_default, help=acpsnimin_helpstr) acpsnew_choices = ["strength", "persistence"] acpsnew_default = "strength" acpsnew_helpstr = \ f"Edge weighting method for the acPSN (default: " \ f"{acpsnew_default})" parser.add_argument("--acpsn-ew", "--acpsn-edge-weights", action="store", type=str, dest="acpsn_ew", choices=acpsnew_choices, default=acpsnew_default, help=acpsnew_helpstr) acpsnnffile_default = \ pkg_resources.resource_filename("pyinteraph", "normalization_factors.ini") acpsnnffile_helpstr = \ f"File with normalization factors to be used " \ f"in the calculation of the acPSN (default: " \ f"{acpsnnffile_default})" parser.add_argument("--acpsn-nf-file", action="store", type=str, dest="nf_file", default=acpsnnffile_default, help=acpsnnffile_helpstr) acpsnnf_default = 999.9 acpsnnfpermissive_helpstr = \ f"Permissive mode. If a residue with no associated " \ f"normalization factor is found, the default normalization " \ f"factor {acpsnnf_default} will be used, and no error will " \ f"be thrown" parser.add_argument("--acpsn-nf-permissive", action="store_true", dest="nf_permissive", help=acpsnnfpermissive_helpstr) acpsnnf_helpstr = \ f"Default normalization factor to be used when running in " \ f"permissive mode (default: {acpsnnf_default})" parser.add_argument("--acpsn-nf-default", action="store", type=str, dest="nf_default", default=acpsnnf_default, help=acpsnnf_helpstr) acpsncsv_default = "acpsn.csv" acpsncsv_helpstr = \ f"Name of the CSV file where to store the list of contacts " \ f"found in the acPSN (default: {acpsncsv_default})" parser.add_argument("--acpsn-csv", action="store", type=str, dest="acpsn_csv", default=acpsncsv_default, help=acpsncsv_helpstr) acpsngraph_helpstr = \ "File where to store the adjacency matrix for the " \ "interaction graph (acPSN)" parser.add_argument("--acpsn-graph", action="store", dest="acpsn_graph", type=str, default=None, help=acpsngraph_helpstr) #---------------------- Hydrophobic contacts ---------------------# hcco_default = 5.0 hcco_helpstr = \ f"Distance cut-off for hydrophobic contacts (default: " \ f"{hcco_default})" parser.add_argument("--hc-co", "--hc-cutoff", action="store", dest="hc_co", type=float, default=hcco_default, help=hcco_helpstr) hcperco_default = 0.0 hcperco_helpstr = \ f"Minimum persistence for hydrophobic contacts (default: " \ f"{hcperco_default})" parser.add_argument("--hc-perco", "--hc-persistence-cutoff", action="store", type=float, dest="hc_perco", default=hcperco_default, help=hcperco_helpstr) hc_reslist = \ ["ALA", "VAL", "LEU", "ILE", "PHE", "PRO", "TRP", "MET"] hcres_helpstr = \ f"Comma-separated list of hydrophobic residues (default: " \ f"{', '.join(hc_reslist)})" parser.add_argument("--hc-residues", action="store", type=str, dest="hc_reslist", default=hc_reslist, help=hcres_helpstr) hccsv_default = "hydrophobic-clusters.csv" hccsv_helpstr = \ f"Name of the CSV file where to store the list of " \ f"hydrophobic contacts found (default: {hccsv_default})" parser.add_argument("--hc-csv", action="store", type=str, dest="hc_csv", default=hccsv_default, help=hccsv_helpstr) hcgraph_helpstr = \ "File where to store the adjacency matrix for the " \ "interaction graph (hydrophobic contacts)" parser.add_argument("--hc-graph", action="store", dest="hc_graph", type=str, default=None, help=hcgraph_helpstr) #-------------------------- Salt bridges -------------------------# sbco_default = 4.5 sbco_helpstr = \ f"Distance cut-off for salt bridges (default: {sbco_default})" parser.add_argument("--sb-co", "--sb-cutoff", action="store", type=float, dest="sb_co", default=sbco_default, help=sbco_helpstr) sbperco_default = 0.0 sbperco_helpstr = \ f"Minimum persistence for salt bridges (default: " \ f"{sbperco_default})" parser.add_argument("--sb-perco", "--sb-persistence-cutoff", action="store", type=float, dest="sb_perco", default=sbperco_default, help=sbperco_helpstr) sbmode_choices = ["different_charge", "same_charge", "all"] sbmode_default = "different_charge" sbmode_helpstr = \ f"Electrostatic interactions mode (default: {sbmode_default})" parser.add_argument("--sb-mode", action="store", type=str, dest="sb_mode", choices=sbmode_choices, default=sbmode_default, help=sbmode_helpstr) sbcgfile_default = \ pkg_resources.resource_filename("pyinteraph", "charged_groups.ini") sbcgfile_helpstr = \ f"File with charged groups to be used to find " \ f"salt bridges (default: {sbcgfile_default})" parser.add_argument("--sb-cg-file", action="store", type=str, dest="cgs_file", default=sbcgfile_default, help=sbcgfile_helpstr) sbcsv_default = "salt-bridges.csv" sbcsv_helpstr = \ f"Name of the CSV file where to store the list of " \ f"salt bridges found (default: {sbcsv_default})" parser.add_argument("--sb-csv", action="store", type=str, dest="sb_csv", default=sbcsv_default, help=sbcsv_helpstr) sbgraph_helpstr = \ "File where to store the adjacency matrix for the " \ "interaction graph (salt bridges)" parser.add_argument("--sb-graph", action="store", type=str, dest="sb_graph", default=None, help=sbgraph_helpstr) #------------------------- Hydrogen bonds ------------------------# hbco_default = 3.5 hbco_helpstr = \ f"Donor-acceptor distance cut-off for hydrogen bonds " \ f"(default: {hbco_default})" parser.add_argument("--hb-co", "--hb-cutoff", action="store", type=float, dest="hb_co", default=hbco_default, help=hbco_helpstr) hbang_default = 120.0 hbang_helpstr = \ f"Donor-acceptor angle cut-off for hydrogen bonds " \ f"(default: {hbang_default})" parser.add_argument("--hb-ang", "--hb-angle", action="store", type=float, dest="hb_angle", default=hbang_default, help=hbang_helpstr) hbperco_default = 0.0 hbperco_helpstr = \ f"Minimum persistence for hydrogen bonds (default: " \ f"{hbperco_default})" parser.add_argument("--hb-perco", action="store", type=float, dest="hb_perco", default=hbperco_default, help=hbperco_helpstr) hbclass_choices = ["all", "mc-mc", "mc-sc", "sc-sc", "custom"] hbclass_default = "all" hbclass_helpstr = \ f"Class of hydrogen bonds to analyze (default: " \ f"{hbclass_default})" parser.add_argument("--hb-class", action="store", type=str, dest="hb_class", choices=hbclass_choices, default=hbclass_default, help=hbclass_helpstr) hbcustom1_helpstr = "Custom group 1 for hydrogen bonds calculation" parser.add_argument("--hb-custom-group-1", action="store", type=str, dest="hb_group1", default=None, help=hbcustom1_helpstr) hbcustom2_helpstr = "Custom group 2 for hydrogen bonds calculation" parser.add_argument("--hb-custom-group-2", action="store", type=str, dest="hb_group2", default=None, help=hbcustom2_helpstr) hbadfile_default = \ pkg_resources.resource_filename("pyinteraph", "hydrogen_bonds.ini") hbadfile_helpstr = \ f"File defining hydrogen bonds donor and acceptor atoms " \ f"(default: {hbadfile_default})" parser.add_argument("--hb-ad-file", action="store", type=str, dest="hbs_file", default=hbadfile_default, help=hbadfile_helpstr) hbcsv_default = "hydrogen-bonds.csv" hbcsv_helpstr = \ f"Name of the CSV file where to store the list of hydrogen " \ f"bonds found (default: {hbcsv_default})" parser.add_argument("--hb-csv", action="store", type=str, dest="hb_csv", default=hbcsv_default, help=hbcsv_helpstr) hbgraph_helpstr = \ "File where to store the adjacency matrix for the " \ "interaction graph (hydrogen bonds)" parser.add_argument("--hb-graph", action="store", type=str, dest="hb_graph", default=None, help=hbgraph_helpstr) #--------------------------- Potential ---------------------------# kbpkbt_default = 1.0 kbpkbt_helpstr = \ f"kb*T value used in the inverse-Boltzmann relation for the " \ f"knowledge-based potential (default: {kbpkbt_default})" parser.add_argument("--kbp-kbt", action="store", type=float, dest="kbp_kbt", default=kbpkbt_default, help=kbpkbt_helpstr) kbpff_default = \ pkg_resources.resource_filename("pyinteraph", "ff.S050.bin64") kbpff_helpstr = \ f"Statistical potential definition file (default: " \ f"{hbperco_default})" parser.add_argument("--kbp-ff", "--force-field", action="store", type=str, dest="kbp_ff", default=kbpff_default, help=kbpff_helpstr) kbpatom_default = \ pkg_resources.resource_filename("pyinteraph", "kbp_atomlist") kbpatom_helpstr = \ f"Ordered, force-field specific list of atom names " \ f"(default: {kbpatom_default})" parser.add_argument("--kbp-atomlist", action="store", type=str, dest="kbp_atomlist", default=kbpatom_default, help=kbpatom_helpstr) kbpcsv_default = "kb-potential.csv" kbpcsv_helpstr = \ f"File where to store the results of the statistical " \ f"potential analysis (default: {kbpcsv_default})" parser.add_argument("--kbp-csv", action="store", type=str, dest="kbp_csv", default=kbpcsv_default, help=kbpcsv_helpstr) kbpgraph_helpstr = \ "File where to store the adjacency matrix for the " \ "interaction graph (statistical potential)" parser.add_argument("--kbp-graph", action="store", type=str, dest="kbp_graph", default=None, help=kbpgraph_helpstr) #-------------------------- Miscellanea --------------------------# ffmasses_dir = "ff_masses" ffmasses_dir = \ pkg_resources.resource_filename("pyinteraph", ffmasses_dir) ffmasses_choices = os.listdir(ffmasses_dir) ffmasses_default = "charmm27" ffmasses_helpstr = \ f"Force field to be used (for masses calculation only) " \ f"(default: {ffmasses_default})" parser.add_argument("--ff-masses", action="store", type=str, dest="ffmasses", choices=ffmasses_choices, default=ffmasses_default, help=ffmasses_helpstr) v_helpstr = "Verbose mode" parser.add_argument("-v", "--verbose", action="store_true", dest="verbose", help=v_helpstr) args = parser.parse_args() ########################## LOGGER SETUP ########################### # Logging format LOGFMT = "%(levelname)s: %(message)s" # Verbose mode? if args.verbose: log.basicConfig(level=log.INFO, format=LOGFMT) else: log.basicConfig(level=log.WARNING, format=LOGFMT) ############################ ARGUMENTS ############################ # Input files top = args.top trj = args.trj ref = args.ref # cmPSN do_cmpsn = args.do_cmpsn cmpsn_co = args.cmpsn_co cmpsn_perco = args.cmpsn_perco if type(args.cmpsn_reslist) is str: cmpsn_reslist = [l.strip() for l in args.cmpsn_reslist.split(",")] else: cmpsn_reslist = args.cmpsn_reslist cmpsn_correction = args.cmpsn_correction cmpsn_csv = args.cmpsn_csv cmpsn_graph = args.cmpsn_graph # acPSN do_acpsn = args.do_acpsn acpsn_co = args.acpsn_co acpsn_perco = args.acpsn_perco acpsn_proxco = args.acpsn_proxco acpsn_imin = args.acpsn_imin acpsn_ew = args.acpsn_ew nf_file = args.nf_file nf_permissive = args.nf_permissive nf_default = args.nf_default acpsn_csv = args.acpsn_csv acpsn_graph = args.acpsn_graph # Hydrophobic contacts do_hc = args.do_hc hc_co = args.hc_co hc_perco = args.hc_perco if type(args.hc_reslist) is str: hc_reslist = [l.strip() for l in args.hc_reslist.split(",")] else: hc_reslist = args.hc_reslist hc_csv = args.hc_csv hc_graph = args.hc_graph # Salt bridges do_sb = args.do_sb sb_co = args.sb_co sb_perco = args.sb_perco sb_mode = args.sb_mode cgs_file = args.cgs_file sb_csv = args.sb_csv sb_graph = args.sb_graph # Hydrogen bonds do_hb = args.do_hb hb_angle = args.hb_angle hb_co = args.hb_co hb_perco = args.hb_perco hb_class = args.hb_class hb_group1 = args.hb_group1 hb_group2 = args.hb_group2 hbs_file = args.hbs_file hb_csv = args.hb_csv hb_graph = args.hb_graph # Statistical potential do_kbp = args.do_kbp kbp_kbt = args.kbp_kbt kbp_atomlist = args.kbp_atomlist kbp_ff = args.kbp_ff kbp_csv = args.kbp_csv kbp_graph = args.kbp_graph # Miscellanea ffmasses = os.path.join(ffmasses_dir, args.ffmasses) ########################## CHECK INPUTS ########################### # Topology and trajectory must be present if not top or not trj: log.error("Topology and trajectory are required.") exit(1) # If no reference structure is passed, the topology will be # the reference if not ref: ref = top log.info("Using topology as reference structure.") # Load systems try: pdb = mda.Universe(ref) uni = mda.Universe(top, trj) except ValueError: logstr = \ "Could not read one of the input files, or trajectory " \ "and topology are not compatible." log.error(logstr) exit(1) ###################### HYDROPHOBIC CONTACTS ####################### if do_hc: # Function to compute the full matrix hc_fmfunc = \ None if hc_graph is None else li.calc_sc_fullmatrix # Compute the table and the matrix hc_table_out, hc_mat_out = \ li.do_interact(identfunc = li.generate_sc_identifiers, pdb = pdb, uni = uni, co = hc_co, perco = hc_perco, ffmasses = ffmasses, fullmatrixfunc = hc_fmfunc, mindist = False, reslist = hc_reslist, correction_func = li.null_correction) # Save .csv hc_table_dict = li.create_dict_tables(hc_table_out) li.save_output_dict(hc_table_dict, hc_csv) # Save .dat (if available) (hc_graph being not None has # been checked already) if hc_mat_out is not None: hc_mat_dict = \ li.create_dict_matrices(hc_mat_out, hc_table_dict, pdb) li.save_output_dict(hc_mat_dict, hc_graph) ############################## cmPSN ############################## if do_cmpsn: # Function to compute the full matrix cmpsn_fmfunc = \ None if not cmpsn_graph else li.calc_sc_fullmatrix # Select the appropriate correction function if cmpsn_correction == "null": selected_func = li.null_correction elif cmpsn_correction == "rg": selected_func = li.rg_correction # Compute the table and the matrix cmpsn_table_out, cmpsn_mat_out = \ li.do_interact(identfunc = li.generate_sc_identifiers, pdb = pdb, uni = uni, co = cmpsn_co, perco = cmpsn_perco, ffmasses = ffmasses, fullmatrixfunc = cmpsn_fmfunc, mindist = False, reslist = cmpsn_reslist, correction_func = selected_func) # Save .csv cmpsn_table_dict = li.create_dict_tables(cmpsn_table_out) li.save_output_dict(cmpsn_table_dict, cmpsn_csv) # Save .dat (if available) if cmpsn_mat_out is not None and cmpsn_graph is not None: cmpsn_mat_dict = \ li.create_dict_matrices(cmpsn_mat_out, cmpsn_table_dict, pdb) li.save_output_dict(cmpsn_mat_dict, cmpsn_graph) ############################## acPSN ############################## if do_acpsn: # Try to parse the charged groups definitions file try: norm_facts = li.parse_nf_file(nf_file) except IOError: logstr = f"Problems reading file {nf_file}." log.error(logstr, exc_info=True) exit(1) except: logstr = \ f"Could not parse the normalization factors file " \ f"{nf_file}. Are there any inconsistencies?" log.error(logstr, exc_info=True) exit(1) # Try to compute the table of contacts and the matrix try: acpsn_table_out, acpsn_mat_out = \ li.do_acpsn(pdb = pdb, uni = uni, co = acpsn_co, perco = acpsn_perco, proxco = acpsn_proxco, imin = acpsn_imin, edge_weight = acpsn_ew, norm_facts = norm_facts, nf_permissive = nf_permissive, nf_default = nf_default) # In case something went wront, report it and exit except Exception as e: logstr = \ f"Could not compute the table of contacts and the " \ f"matrix for acPSN: {e}" log.error(logstr) exit(1) # Save .csv acpsn_table_dict = li.create_dict_tables(acpsn_table_out) li.save_output_dict(acpsn_table_dict, acpsn_csv) # Save .dat (if available) if acpsn_mat_out is not None and acpsn_graph is not None: acpsn_mat_dict = \ li.create_dict_matrices(acpsn_mat_out, acpsn_table_dict, pdb) li.save_output_dict(acpsn_mat_dict, acpsn_graph) ########################## SALT BRIDGES ########################### if do_sb: # Try to parse the charged groups definitions file try: cgs = li.parse_cgs_file(cgs_file) except IOError: logstr = f"Problems reading file {cgs_file}." log.error(logstr, exc_info=True) exit(1) except: logstr = \ f"Could not parse the charged groups file " \ f"{cgs_file}. Are there any inconsistencies?" log.error(logstr, exc_info=True) exit(1) if sb_mode == "same_charge": sb_mode = "same" elif sb_mode == "different_charge": sb_mode = "diff" elif sb_mode == "all": sb_mode = "both" # Function to compute the full matrix sb_fmfunc = None if not sb_graph else li.calc_cg_fullmatrix # Compute the table and the matrix sb_table_out, sb_mat_out = \ li.do_interact(identfunc = li.generate_cg_identifiers, pdb = pdb, uni = uni, co = sb_co, perco = sb_perco, ffmasses = ffmasses, fullmatrixfunc = sb_fmfunc, mindist = True, mindist_mode = sb_mode, cgs = cgs) # Save .csv sb_table_dict = li.create_dict_tables(sb_table_out) li.save_output_dict(sb_table_dict, sb_csv) # Save .dat (if available) if sb_mat_out is not None and sb_graph is not None: sb_mat_dict = \ li.create_dict_matrices(sb_mat_out, sb_table_dict, pdb) li.save_output_dict(sb_mat_dict, sb_graph) ########################### HYDROGEN BONDS ############################ if do_hb: # Atom selection for main chain hydrogen bonds mc_sel = "backbone or name H or name H1 or name H2 " \ "or name H3 or name O1 or name O2 or name OXT" # Atom selection for side chain hydrogen bonds sc_sel = f"protein and (not {mc_sel})" # Custom atom selection if (hb_group1 and hb_group2) and hb_class != "custom": warnstr = \ "Hydrogen bond custom groups have been specified; " \ "they will be used. Please use --hb-class=custom to " \ "get rid of this warning!" log.warning(warnstr) hb_class = "custom" if hb_class == "custom": if not hb_group1 or not hb_group2: errstr = \ "Hydrogen bond class 'custom' requires the " \ "definition of two interation groups. (see " \ "options --hb-custom-group1 and --hb-custom-group2)" log.error(errstr) exit(1) # All hydrogen bonds elif hb_class == "all": hb_group1 = "protein" hb_group2 = "protein" # Main chain - main chain hydrogen bonds elif hb_class == "mc-mc": hb_group1 = mc_sel hb_group2 = mc_sel # Side chain - side chain hydrogen bonds elif hb_class == "sc-sc": hb_group1 = sc_sel hb_group2 = sc_sel # Main chain - side chain hydrogen bonds elif hb_class == "mc-sc": hb_group1 = mc_sel hb_group2 = sc_sel # Check if selection 1 is valid try: uni.select_atoms(hb_group1) except: log.error("Selection 1 is invalid", exc_info=True) exit(1) # Check if selection 2 is valid try: uni.select_atoms(hb_group2) except: log.error("Selection 2 is invalid", exc_info=True) exit(1) # Check the donors-acceptors file try: hbs = li.parse_hbs_file(hbs_file) except IOError: logstr = f"Problems reading {hbs_file}." log.error(logstr, exc_info=True) exit(1) except: logstr = f"Could not parse {hbs_file}." log.error(logstr, exc_info=True) exit(1) # Whether to compute the full matrix do_fullmatrix = True if hb_graph else False # Whtether to group the hydrogen bonds per residue perresidue = False # Compute the table and the matrix hb_table_out, hb_mat_out = \ li.do_hbonds(sel1 = hb_group1, sel2 = hb_group2, pdb = pdb, uni = uni, distance = hb_co, angle = hb_angle, perco = hb_perco, do_fullmatrix = do_fullmatrix, other_hbs = hbs, perresidue = perresidue) # Save .csv hb_table_dict = li.create_dict_tables(hb_table_out) li.save_output_dict(hb_table_dict, hb_csv) # Save .dat (if available) if hb_mat_out is not None and hb_graph is not None: hb_mat_dict = \ li.create_dict_matrices(hb_mat_out, hb_table_dict, pdb) li.save_output_dict(hb_mat_dict, hb_graph) ######################## STATISTICAL POTENTIAL ######################## if do_kbp: # Residue list for potential calculation - all canonical but GLY kbp_reslist = \ ["ALA", "ARG", "ASN", "ASP", "CYS", "GLN", "GLU", "HIS", "ILE", "LEU", "LYS", "MET", "PHE", "PRO", "SER", "THR", "TRP", "TYR", "VAL"] # Parse the atom list kbp_atomlist = li.parse_atomlist(kbp_atomlist) # Whether to compute the full matrix do_fullmatrix = True if kbp_graph else False # Compute the output table and the matrix kbp_table_out, kbp_mat_out = \ li.do_potential(kbp_atomlist = kbp_atomlist, residues_list = kbp_reslist, potential_file = kbp_ff, uni = uni, pdb = pdb, do_fullmatrix = do_fullmatrix, kbT = kbp_kbt, seq_dist_co = 0) # Save .csv kbp_table_dict = li.create_dict_tables(kbp_table_out) li.save_output_dict(kbp_table_dict, kbp_csv) # Save .mat (if available) if kbp_mat_out is not None and kbp_graph is not None: np.savetxt(kbp_graph, kbp_mat_out, fmt="%.3f")