pmi/io/__init__.py

"""@namespace IMP.pmi.io
   Utility classes and functions for reading and storing PMI files
"""

import IMP
import IMP.algebra
import IMP.atom
import IMP.pmi
import IMP.rmf
import IMP.pmi.analysis
import IMP.pmi.output
import IMP.pmi.tools
import RMF
import os
import numpy as np
from collections import defaultdict
from pathlib import Path


def parse_dssp(dssp_fn, limit_to_chains='', name_map=None):
    """Read a DSSP file, and return secondary structure elements (SSEs).
    Values are all PDB residue numbering.
    @param dssp_fn The file to read
    @param limit_to_chains Only read/return these chain IDs
    @param name_map If passed, return tuples organized by molecule name
           (name_map should be a dictionary with chain IDs as keys and
           molecule names as values).

    @return a dictionary with keys 'helix', 'beta', 'loop'.
            Each contains a list of SSEs.
            Each SSE is a list of elements (e.g. strands in a sheet).
            Each element is a tuple (residue start, residue end, chain).

    Example for a structure with helix A:5-7 and Beta strands A:1-3,A:9-11:

    @code{.py}
    ret = { 'helix' : [ [ (5,7,'A') ], ...]
            'beta'  : [ [ (1,3,'A'),
                          (9,11,'A'), ...], ...]
            'loop'  : same format as helix
          }
    @endcode
    """

    def convert_chain(ch):
        if name_map is None:
            return ch
        else:
            return name_map.get(ch, ch)

    # setup
    helix_classes = 'GHI'
    strand_classes = 'EB'
    loop_classes = [' ', '', 'T', 'S']
    sse_dict = {}
    for h in helix_classes:
        sse_dict[h] = 'helix'
    for s in strand_classes:
        sse_dict[s] = 'beta'
    for lc in loop_classes:
        sse_dict[lc] = 'loop'
    sses = {'helix': [], 'beta': [], 'loop': []}

    # read file and parse
    start = False

    # temporary beta dictionary indexed by DSSP's ID
    beta_dict = IMP.pmi.tools.OrderedDefaultDict(list)
    prev_sstype = None
    prev_beta_id = None

    with open(dssp_fn, 'r') as fh:
        for line in fh:
            fields = line.split()
            chain_break = False
            if len(fields) < 2:
                continue
            if fields[1] == "RESIDUE":
                # Start parsing from here
                start = True
                continue
            if not start:
                continue
            if line[9] == " ":
                chain_break = True
            elif limit_to_chains != '' and line[11] not in limit_to_chains:
                continue

            # gather line info
            if not chain_break:
                pdb_res_num = int(line[5:10])
                chain = line[11]
                sstype = sse_dict[line[16]]
                beta_id = line[33]

            # decide whether to extend or store the SSE
            if prev_sstype is None:
                cur_sse = [pdb_res_num, pdb_res_num, convert_chain(chain)]
            elif sstype != prev_sstype or chain_break:
                # add cur_sse to the right place
                if prev_sstype in ['helix', 'loop']:
                    sses[prev_sstype].append([cur_sse])
                else:  # prev_sstype == 'beta'
                    beta_dict[prev_beta_id].append(cur_sse)
                cur_sse = [pdb_res_num, pdb_res_num, convert_chain(chain)]
            else:
                cur_sse[1] = pdb_res_num
            if chain_break:
                prev_sstype = None
                prev_beta_id = None
            else:
                prev_sstype = sstype
                prev_beta_id = beta_id

    # final SSE processing
    if prev_sstype in ['helix', 'loop']:
        sses[prev_sstype].append([cur_sse])
    elif prev_sstype == 'beta':
        beta_dict[prev_beta_id].append(cur_sse)
    # gather betas
    for beta_sheet in beta_dict:
        sses['beta'].append(beta_dict[beta_sheet])
    return sses


def save_best_models(model, out_dir, stat_files,
                     number_of_best_scoring_models=10, get_every=1,
                     score_key="SimplifiedModel_Total_Score_None",
                     feature_keys=None, rmf_file_key="rmf_file",
                     rmf_file_frame_key="rmf_frame_index",
                     override_rmf_dir=None):
    """Given a list of stat files, read them all and find the best models.
    Save to a single RMF along with a stat file.
    @param model The IMP Model
    @param out_dir The output directory. Will save 3 files (RMF, stat, summary)
    @param stat_files List of all stat files to collect
    @param number_of_best_scoring_models Num best models to gather
    @param get_every Skip frames
    @param score_key Used for the ranking
    @param feature_keys Keys to keep around
    @param rmf_file_key The key that says RMF file name
    @param rmf_file_frame_key The key that says RMF frame number
    @param override_rmf_dir For output, change the name of the RMF
           directory (experiment)
    """

    # start by splitting into jobs
    try:
        from mpi4py import MPI
        comm = MPI.COMM_WORLD
        rank = comm.Get_rank()
        number_of_processes = comm.size
    except ImportError:
        rank = 0
        number_of_processes = 1
    my_stat_files = IMP.pmi.tools.chunk_list_into_segments(
        stat_files, number_of_processes)[rank]

    # filenames
    out_dir = Path(out_dir)
    out_stat_fn = out_dir / ("top_%d.out" % number_of_best_scoring_models)
    out_rmf_fn = out_dir / ("top_%d.rmf3" % number_of_best_scoring_models)

    # extract all the models
    all_fields = []
    for nsf, sf in enumerate(my_stat_files):

        # get list of keywords
        root_directory_of_stat_file = Path(sf).parent.parent
        print("getting data from file %s" % sf)
        po = IMP.pmi.output.ProcessOutput(sf)
        all_keys = [score_key,
                    rmf_file_key,
                    rmf_file_frame_key]
        for k in po.get_keys():
            for fk in feature_keys:
                if fk in k:
                    all_keys.append(k)
        fields = po.get_fields(all_keys,
                               get_every=get_every)

        # check that all lengths are all equal
        length_set = set([len(fields[f]) for f in fields])
        minlen = min(length_set)
        # if some of the fields are missing, truncate
        # the feature files to the shortest one
        if len(length_set) > 1:
            print("get_best_models: the statfile is not synchronous")
            minlen = min(length_set)
            for f in fields:
                fields[f] = fields[f][0:minlen]
        if nsf == 0:
            all_fields = fields
        else:
            for k in fields:
                all_fields[k] += fields[k]

        if override_rmf_dir is not None:
            for i in range(minlen):
                all_fields[rmf_file_key][i] = os.path.join(
                    override_rmf_dir,
                    os.path.basename(all_fields[rmf_file_key][i]))

    # gather info, sort, write
    if number_of_processes != 1:
        comm.Barrier()
    if rank != 0:
        comm.send(all_fields, dest=0, tag=11)
    else:
        for i in range(1, number_of_processes):
            data_tmp = comm.recv(source=i, tag=11)
            for k in all_fields:
                all_fields[k] += data_tmp[k]

        # sort by total score
        order = sorted(range(len(all_fields[score_key])),
                       key=lambda i: float(all_fields[score_key][i]))

        # write the stat and RMF files
        stat = open(str(out_stat_fn), 'w')
        rh0 = RMF.open_rmf_file_read_only(
            str(root_directory_of_stat_file / all_fields[rmf_file_key][0]))
        prots = IMP.rmf.create_hierarchies(rh0, model)
        del rh0
        outf = RMF.create_rmf_file(str(out_rmf_fn))
        IMP.rmf.add_hierarchies(outf, prots)
        for nm, i in enumerate(order[:number_of_best_scoring_models]):
            dline = dict((k, all_fields[k][i]) for k in all_fields)
            dline['orig_rmf_file'] = dline[rmf_file_key]
            dline['orig_rmf_frame_index'] = dline[rmf_file_frame_key]
            dline[rmf_file_key] = str(out_rmf_fn)
            dline[rmf_file_frame_key] = nm
            rh = RMF.open_rmf_file_read_only(
                str(root_directory_of_stat_file / all_fields[rmf_file_key][i]))
            IMP.rmf.link_hierarchies(rh, prots)
            IMP.rmf.load_frame(rh,
                               RMF.FrameID(all_fields[rmf_file_frame_key][i]))
            IMP.rmf.save_frame(outf)
            del rh
            stat.write(str(dline) + '\n')
        stat.close()
        print('wrote stats to', out_stat_fn)
        print('wrote rmfs to', out_rmf_fn)


class _TempProvenance:
    """Placeholder to track provenance information added to the IMP model.
       This is since we typically don't preserve the IMP::Model object
       throughout a PMI protocol."""
    def __init__(self, provclass, particle_name, *args, **kwargs):
        self.provclass = provclass
        self.particle_name = particle_name
        self.args = args
        self.kwargs = kwargs

    def get_decorator(self, model):
        """Instantiate and return an IMP Provenance object in the model."""
        pi = model.add_particle(self.particle_name)
        return self.provclass.setup_particle(model, pi, *self.args,
                                             **self.kwargs)


class ClusterProvenance(_TempProvenance):
    def __init__(self, *args, **kwargs):
        _TempProvenance.__init__(self, IMP.core.ClusterProvenance,
                                 "clustering", *args, **kwargs)


class FilterProvenance(_TempProvenance):
    def __init__(self, *args, **kwargs):
        _TempProvenance.__init__(self, IMP.core.FilterProvenance,
                                 "filtering", *args, **kwargs)


class CombineProvenance(_TempProvenance):
    def __init__(self, *args, **kwargs):
        _TempProvenance.__init__(self, IMP.core.CombineProvenance,
                                 "combine runs", *args, **kwargs)


def add_provenance(prov, hiers):
    """Add provenance information in `prov` (a list of _TempProvenance
       objects) to each of the IMP hierarchies provided.
       Note that we do this all at once since we typically don't preserve
       the IMP::Model object throughout a PMI protocol."""
    for h in hiers:
        IMP.pmi.tools._add_pmi_provenance(h)
        m = h.get_model()
        for p in prov:
            IMP.core.add_provenance(m, h, p.get_decorator(m))


def get_best_models(stat_files,
                    score_key="SimplifiedModel_Total_Score_None",
                    feature_keys=None,
                    rmf_file_key="rmf_file",
                    rmf_file_frame_key="rmf_frame_index",
                    prefiltervalue=None,
                    get_every=1, provenance=None):
    """ Given a list of stat files, read them all and find the best models.
    Returns the best rmf filenames, frame numbers, scores, and values
    for feature keywords
    """
    rmf_file_list = []              # best RMF files
    rmf_file_frame_list = []        # best RMF frames
    score_list = []                 # best scores
    # best values of the feature keys
    feature_keyword_list_dict = defaultdict(list)
    statistics = IMP.pmi.output.OutputStatistics()
    for sf in stat_files:
        root_directory_of_stat_file = os.path.dirname(os.path.abspath(sf))
        if sf[-4:] == 'rmf3':
            root_directory_of_stat_file = os.path.dirname(
                os.path.abspath(root_directory_of_stat_file))
        print("getting data from file %s" % sf)
        po = IMP.pmi.output.ProcessOutput(sf)

        try:
            file_keywords = po.get_keys()
        except:  # noqa: E722
            continue

        keywords = [score_key, rmf_file_key, rmf_file_frame_key]

        # check all requested keys are in the file
        #  this looks weird because searching for "*requested_key*"
        if feature_keys:
            for requested_key in feature_keys:
                for file_k in file_keywords:
                    if requested_key in file_k:
                        if file_k not in keywords:
                            keywords.append(file_k)

        if prefiltervalue is None:
            fields = po.get_fields(keywords,
                                   get_every=get_every,
                                   statistics=statistics)
        else:
            fields = po.get_fields(
                keywords, filtertuple=(score_key, "<", prefiltervalue),
                get_every=get_every, statistics=statistics)

        # check that all lengths are all equal
        length_set = set()
        for f in fields:
            length_set.add(len(fields[f]))

        # if some of the fields are missing, truncate
        # the feature files to the shortest one
        if len(length_set) > 1:
            print("get_best_models: the statfile is not synchronous")
            minlen = min(length_set)
            for f in fields:
                fields[f] = fields[f][0:minlen]

        # append to the lists
        score_list += fields[score_key]
        for rmf in fields[rmf_file_key]:
            rmf = os.path.normpath(rmf)
            if root_directory_of_stat_file not in rmf:
                rmf_local_path = os.path.join(
                    os.path.basename(os.path.dirname(rmf)),
                    os.path.basename(rmf))
                rmf = os.path.join(root_directory_of_stat_file, rmf_local_path)
            rmf_file_list.append(rmf)

        rmf_file_frame_list += fields[rmf_file_frame_key]

        for k in keywords:
            feature_keyword_list_dict[k] += fields[k]

    # Record combining and filtering operations in provenance, if requested
    if provenance is not None:
        if len(stat_files) > 1:
            provenance.append(CombineProvenance(len(stat_files),
                                                statistics.total))
        if get_every != 1:
            provenance.append(
                FilterProvenance("Keep fraction", 0.,
                                 statistics.passed_get_every))
        if prefiltervalue is not None:
            provenance.append(FilterProvenance("Total score",
                                               prefiltervalue,
                                               statistics.passed_filtertuple))

    return (rmf_file_list, rmf_file_frame_list, score_list,
            feature_keyword_list_dict)


def get_trajectory_models(stat_files,
                          score_key="SimplifiedModel_Total_Score_None",
                          rmf_file_key="rmf_file",
                          rmf_file_frame_key="rmf_frame_index",
                          get_every=1):
    """Given a list of stat files, read them all and find a trajectory
       of models. Returns the rmf filenames, frame numbers, scores, and
       values for feature keywords
    """
    rmf_file_list = []              # best RMF files
    rmf_file_frame_list = []        # best RMF frames
    score_list = []                 # best scores
    for sf in stat_files:
        root_directory_of_stat_file = Path(sf).parent.parent
        print("getting data from file %s" % sf)
        po = IMP.pmi.output.ProcessOutput(sf)

        feature_keywords = [score_key, rmf_file_key, rmf_file_frame_key]

        fields = po.get_fields(feature_keywords, get_every=get_every)

        # check that all lengths are all equal
        length_set = set()
        for f in fields:
            length_set.add(len(fields[f]))

        # if some of the fields are missing, truncate
        # the feature files to the shortest one
        if len(length_set) > 1:
            print("get_best_models: the statfile is not synchronous")
            minlen = min(length_set)
            for f in fields:
                fields[f] = fields[f][0:minlen]

        # append to the lists
        score_list += fields[score_key]
        for rmf in fields[rmf_file_key]:
            rmf_file_list.append(str(root_directory_of_stat_file / rmf))

        rmf_file_frame_list += fields[rmf_file_frame_key]

    return rmf_file_list, rmf_file_frame_list, score_list


def read_coordinates_of_rmfs(model,
                             rmf_tuples,
                             alignment_components=None,
                             rmsd_calculation_components=None,
                             state_number=0):
    """Read in coordinates of a set of RMF tuples.
    Returns the coordinates split as requested (all, alignment only, rmsd only)
    as well as RMF file names (as keys in a dictionary, with values being
    the rank number) and just a plain list
    @param model      The IMP model
    @param rmf_tuples [score,filename,frame number,original order number, rank]
    @param alignment_components Tuples to specify what you're aligning on
    @param rmsd_calculation_components Tuples to specify what components
           are used for RMSD calc
    @param state_number The state to extract from the file
    """
    all_coordinates = []
    rmsd_coordinates = []
    alignment_coordinates = []
    all_rmf_file_names = []
    rmf_file_name_index_dict = {}  # storing the features

    for cnt, tpl in enumerate(rmf_tuples):
        rmf_file = tpl[1]
        frame_number = tpl[2]
        if cnt == 0:
            prots = IMP.pmi.analysis.get_hiers_from_rmf(model,
                                                        frame_number,
                                                        rmf_file)
        else:
            IMP.pmi.analysis.link_hiers_to_rmf(model, prots, frame_number,
                                               rmf_file)

        if not prots:
            continue
        states = IMP.atom.get_by_type(prots[0], IMP.atom.STATE_TYPE)
        if states:
            prot = states[state_number]
        else:
            # If no states, assume PMI1-style
            prot = prots[state_number]

        # getting the particles
        part_dict = IMP.pmi.analysis.get_particles_at_resolution_one(prot)
        model_coordinate_dict = {}
        template_coordinate_dict = {}
        rmsd_coordinate_dict = {}

        for pr in part_dict:
            model_coordinate_dict[pr] = np.array(
                [np.array(IMP.core.XYZ(i).get_coordinates())
                 for i in part_dict[pr]])
        # for each file, get (as floats) a list of all coordinates
        #  of all requested tuples, organized as dictionaries.
        for tuple_dict, result_dict in zip(
                (alignment_components, rmsd_calculation_components),
                (template_coordinate_dict, rmsd_coordinate_dict)):

            if tuple_dict is None:
                continue

            # PMI2: do selection of resolution and name at the same time
            for pr in tuple_dict:
                ps = IMP.pmi.tools.select_by_tuple_2(
                    prot, tuple_dict[pr], resolution=1)
                result_dict[pr] = [
                    list(map(float, IMP.core.XYZ(p).get_coordinates()))
                    for p in ps]

        all_coordinates.append(model_coordinate_dict)
        alignment_coordinates.append(template_coordinate_dict)
        rmsd_coordinates.append(rmsd_coordinate_dict)
        frame_name = rmf_file + '|' + str(frame_number)
        all_rmf_file_names.append(frame_name)
        rmf_file_name_index_dict[frame_name] = tpl[4]
    return (all_coordinates, alignment_coordinates, rmsd_coordinates,
            rmf_file_name_index_dict, all_rmf_file_names)


def get_bead_sizes(model, rmf_tuple, rmsd_calculation_components=None,
                   state_number=0):
    '''
    @param model      The IMP model
    @param rmf_tuple  score,filename,frame number,original order number, rank
    @param rmsd_calculation_components Tuples to specify what components
           are used for RMSD calc
    @param state_number The state to extract from the file
    '''
    if rmsd_calculation_components is None:
        return {}

    rmf_file = rmf_tuple[1]
    frame_number = rmf_tuple[2]
    prots = IMP.pmi.analysis.get_hiers_from_rmf(model,
                                                frame_number,
                                                rmf_file)

    states = IMP.atom.get_by_type(prots[0], IMP.atom.STATE_TYPE)
    prot = states[state_number]

    rmsd_bead_size_dict = {}

    # PMI2: do selection of resolution and name at the same time
    for pr in rmsd_calculation_components:
        ps = IMP.pmi.tools.select_by_tuple_2(
            prot, rmsd_calculation_components[pr], resolution=1)
        rmsd_bead_size_dict[pr] = [
            len(IMP.pmi.tools.get_residue_indexes(p)) for p in ps]

    return rmsd_bead_size_dict


class TotalScoreOutput:
    """A helper output for model evaluation"""
    def __init__(self, model):
        self.model = model
        self.rs = IMP.pmi.tools.get_restraint_set(self.model)

    def get_output(self):
        score = self.rs.evaluate(False)
        output = {}
        output["Total_Score"] = str(score)
        return output