""" HHfrag: build a dynamic variable-length fragment library for protein structure prediction with Rosetta AbInitio. """ import os import multiprocessing import csb.apps import csb.apps.hhsearch as hhsearch import csb.bio.io.hhpred import csb.bio.fragments import csb.bio.fragments.rosetta as rosetta import csb.bio.structure import csb.io.tsv import csb.core class ExitCodes(csb.apps.ExitCodes): IO_ERROR = 2 INVALID_DATA = 3 HHSEARCH_FAILURE = 4 NO_OUTPUT = 5 class AppRunner(csb.apps.AppRunner): @property def target(self): return HHfragApp def command_line(self): cmd = csb.apps.ArgHandler(self.program, __doc__) cpu = multiprocessing.cpu_count() cmd.add_scalar_option('hhsearch', 'H', str, 'path to the HHsearch executable', default='hhsearch') cmd.add_scalar_option('database', 'd', str, 'database directory (containing PDBS25.hhm)', required=True) cmd.add_scalar_option('min', 'm', int, 'minimum query segment length', default=6) cmd.add_scalar_option('max', 'M', int, 'maximum query segment length', default=21) cmd.add_scalar_option('step', 's', int, 'query segmentation step', default=3) cmd.add_scalar_option('cpu', 'c', int, 'maximum degree of parallelism', default=cpu) cmd.add_scalar_option('gap-filling', 'g', str, 'path to a fragment file (e.g. CSfrag or Rosetta NNmake), which will be used ' 'to complement low-confidence regions (when specified, a hybrid fragment library be produced)') cmd.add_scalar_option('filtered-filling', 'F', str, 'path to a filtered fragment file (e.g. filtered CSfrag-ments), which will ' 'be mixed with the HHfrag-set and then filtered, resulting in a double-filtered library') cmd.add_boolean_option('filtered-map', 'f', 'make an additional filtered fragment map of centroids and predict torsion angles', default=False) cmd.add_boolean_option('c-alpha', None, 'include also C-alpha vectors in the output', default=False) cmd.add_scalar_option('confidence-threshold', 't', float, 'confidence threshold for gap filling', default=0.7) cmd.add_scalar_option('verbosity', 'v', int, 'verbosity level', default=2) cmd.add_scalar_option('output', 'o', str, 'output directory', default='.') cmd.add_positional_argument('QUERY', str, 'query profile HMM (e.g. created with csb.apps.buildhmm)') return cmd class HHfragApp(csb.apps.Application): def main(self): if not os.path.isdir(self.args.output): HHfragApp.exit('Output directory does not exist', code=ExitCodes.INVALID_DATA, usage=True) if self.args.c_alpha: builder = rosetta.ExtendedOutputBuilder else: builder = rosetta.OutputBuilder try: hhf = HHfrag(self.args.QUERY, self.args.hhsearch, self.args.database, logger=self) output = os.path.join(self.args.output, hhf.query.id) hhf.slice_query(self.args.min, self.args.max, self.args.step, self.args.cpu) frags = hhf.extract_fragments() if len(frags) == 0: HHfragApp.exit('No fragments found!', code=ExitCodes.NO_OUTPUT) fragmap = hhf.build_fragment_map() fragmap.dump(output + '.hhfrags.09', builder) if self.args.filtered_map: fragmap, events = hhf.build_filtered_map() fragmap.dump(output + '.filtered.09', builder) tsv = PredictionBuilder.create(events).product tsv.dump(output + '.centroids.tsv') if self.args.filtered_filling: fragmap, events = hhf.build_hybrid_filtered_map(self.args.filtered_filling) fragmap.dump(output + '.hybrid.filtered.09', builder) tsv = PredictionBuilder.create(events).product tsv.dump(output + '.hybrid.centroids.tsv') if self.args.gap_filling: fragmap = hhf.build_combined_map(self.args.gap_filling, self.args.confidence_threshold) fragmap.dump(output + '.complemented.09', builder) self.log('\nDONE.') except ArgumentIOError as ae: HHfragApp.exit(str(ae), code=ExitCodes.IO_ERROR) except ArgumentError as ae: HHfragApp.exit(str(ae), code=ExitCodes.INVALID_DATA) except csb.io.InvalidCommandError as ose: msg = '{0!s}: {0.program}'.format(ose) HHfragApp.exit(msg, ExitCodes.IO_ERROR) except csb.bio.io.hhpred.HHProfileFormatError as hfe: msg = 'Corrupt HMM: {0!s}'.format(hfe) HHfragApp.exit(msg, code=ExitCodes.INVALID_DATA) except csb.io.ProcessError as pe: message = 'Bad exit code from HHsearch: #{0.code}.\nSTDERR: {0.stderr}\nSTDOUT: {0.stdout}'.format(pe.context) HHfragApp.exit(message, ExitCodes.HHSEARCH_FAILURE) def log(self, message, ending='\n', level=1): if level <= self.args.verbosity: super(HHfragApp, self).log(message, ending) class ArgumentError(ValueError): pass class ArgumentIOError(ArgumentError): pass class InvalidOperationError(ValueError): pass class HHfrag(object): """ The HHfrag dynamic fragment detection protocol. @param query: query HMM path and file name @type query: str @param binary: the HHsearch binary @type binary: str @param database: path to the PDBS25 directory @type database: str @param logger: logging client (needs to have a C{log} method) @type logger: L{Application} """ PDBS = 'pdbs25.hhm' def __init__(self, query, binary, database, logger=None): try: self._query = csb.bio.io.HHProfileParser(query).parse() except IOError as io: raise ArgumentIOError(str(io)) self._hsqs = None self._matches = None self._app = logger self._database = None self._pdbs25 = None self._aligner = None self.database = database self.aligner = hhsearch.HHsearch(binary, self.pdbs25, cpu=2) if self.query.layers.length < 1: raise ArgumentError("Zero-length sequence profile") @property def query(self): return self._query @property def pdbs25(self): return self._pdbs25 @property def database(self): return self._database @database.setter def database(self, value): database = value pdbs25 = os.path.join(value, HHfrag.PDBS) if not os.path.isfile(pdbs25): raise ArgumentError('PDBS25 not found here: ' + pdbs25) self._database = database self._pdbs25 = pdbs25 @property def aligner(self): return self._aligner @aligner.setter def aligner(self, value): if hasattr(value, 'run') and hasattr(value, 'runmany'): self._aligner = value else: raise TypeError(value) def log(self, *a, **ka): if self._app: self._app.log(*a, **ka) def slice_query(self, min=6, max=21, step=3, cpu=None): """ Run the query slicer and collect the optimal query segments. @param min: min segment length @type min: int @param max: max segment length @type max: int @param step: slicing step @type step: int @param cpu: degree of parallelism @type cpu: int @rtype: tuple of L{SliceContext} """ if not 0 < min <= max: raise ArgumentError('min and max must be positive numbers, with max >= min') if not 0 < step: raise ArgumentError('step must be positive number') self.log('\n# Processing profile HMM "{0}"...'.format(self.query.id)) self.log('', level=2) qp = self.query hsqs = [] if not cpu: cpu = max - min + 1 for start in range(1, qp.layers.length - min + 1 + 1, step): self.log('{0:3}. '.format(start), ending='', level=1) probes = [] for end in range(start + min - 1, start + max): if end > qp.layers.length: break context = SliceContext(qp.segment(start, end), start, end) probes.append(context) probes = self.aligner.runmany(probes, workers=cpu) probes.sort() if len(probes) > 0: rep = probes[-1] hsqs.append(rep) self.log('{0.start:3} {0.end:3} ({0.length:2} aa) {0.recurrence:3} hits'.format(rep), level=1) else: self.log(' no hits', level=1) self._hsqs = hsqs return tuple(hsqs) def extract_fragments(self): """ Extract all matching fragment instances, given the list of optimal query slices, generated during the first stage. @rtype: tuple of L{Assignment}s """ if self._hsqs is None: self.slice_query() self.log('\n# Extracting fragments...') fragments = [] for si in self._hsqs: self.log('\nSEGMENT: {0.start:3} {0.end:3} ({0.recurrence})'.format(si), level=2) for hit in si.hits: cn = 0 for chunk in hit.alignment.segments: chunk.qstart = chunk.qstart + si.start - 1 chunk.qend = chunk.qend + si.start - 1 cn += 1 self.log(' {0.id:5} L{0.qstart:3} {0.qend:3} {0.length:2}aa P:{0.probability:5.3f}'.format(chunk), ending='', level=2) sourcefile = os.path.join(self.database, hit.id + '.pdb') if not os.path.isfile(sourcefile): self.log(' missing', level=2) continue source = csb.bio.io.StructureParser(sourcefile).parse().first_chain assert hit.id[-1] == source.id source.compute_torsion() try: fragment = csb.bio.fragments.Assignment(source, chunk.start, chunk.end, chunk.qstart, chunk.qend, probability=chunk.probability) fragments.append(fragment) if cn > 1: self.log(' (chunk #{0})'.format(cn), level=2) else: self.log('', level=2) except csb.bio.structure.Broken3DStructureError: self.log(' corrupt', level=2) continue self._matches = fragments return tuple(fragments) def _plot_lengths(self): self.log('\n {0} ungapped assignments'.format(len(self._matches))) self.log('', level=2) histogram = {} for f in self._matches: histogram[f.length] = histogram.get(f.length, 0) + 1 for length in sorted(histogram): percent = histogram[length] * 100.0 / len(self._matches) bar = '{0:3} |{1} {2:5.1f}%'.format(length, 'o' * int(percent), percent) self.log(bar, level=2) def build_fragment_map(self): """ Build a full Rosetta fragset. @rtype: L{RosettaFragmentMap} """ if self._matches is None: self.extract_fragments() self.log('\n# Building dynamic fragment map...') self._plot_lengths() target = csb.bio.fragments.Target.from_profile(self.query) target.assignall(self._matches) factory = csb.bio.fragments.RosettaFragsetFactory() return factory.make_fragset(target) def _filter_event_handler(self, ri): if ri.gap is True or ri.confident is False: self.log('{0.rank:3}. {0.confidence:5.3f} {0.count:3}'.format(ri), level=2) else: phi = PredictionBuilder.format_angle(ri.torsion.phi) psi = PredictionBuilder.format_angle(ri.torsion.psi) omega = PredictionBuilder.format_angle(ri.torsion.omega) pred = "{0.source_id:5} {0.start:3} {0.end:3} {1} {2} {3}".format(ri.rep, phi, psi, omega) self.log('{0.rank:3}. {0.confidence:5.3f} {0.count:3} {1}'.format(ri, pred), level=2) def build_filtered_map(self): """ Build a filtered fragset of centroids. @return: filtered fragset and a list of residue-wise predictions (centroid and torsion angles) @rtype: L{RosettaFragmentMap}, list of L{ResidueEventInfo} """ if self._matches is None: self.extract_fragments() self.log('\n# Building filtered map...') self.log('\n Confidence Recurrence Representative Phi Psi Omega', level=2) events = [] def logger(ri): events.append(ri) self._filter_event_handler(ri) target = csb.bio.fragments.Target.from_profile(self.query) target.assignall(self._matches) factory = csb.bio.fragments.RosettaFragsetFactory() fragset = factory.make_filtered(target, extend=True, callback=logger) return fragset, events def _merge_event_handler(self, ri): marked = "" if ri.gap is True or ri.confident is False: marked = "*" self.log('{0.rank:3}. {0.confidence:5.3f} {0.count:3} {1:>3}'.format(ri, marked), level=2) def build_combined_map(self, fragfile, threshold=0.7, top=25): """ Build a hybrid map, where low-confidence regions are complemented with the specified filling. @param threshold: confidence threshold @type threshold: float @param fragfile: filling fragset (Rosetta fragment file) @type fragfile: str @return: filtered fragset and a list of residue-wise predictions (centroid and torsion angles) @rtype: L{RosettaFragmentMap}, list of L{ResidueEventInfo} """ if self._matches is None: self.extract_fragments() self.log('\n# Building complemented map...') try: filling = rosetta.RosettaFragmentMap.read(fragfile, top=top) except IOError as io: raise ArgumentIOError(str(io)) self.log('\n {0} supplementary fragments loaded'.format(filling.size)) self.log(' Confidence Recurrence Fill?', level=2) target = csb.bio.fragments.Target.from_profile(self.query) target.assignall(self._matches) factory = csb.bio.fragments.RosettaFragsetFactory() return factory.make_combined(target, filling, threshold=threshold, callback=self._merge_event_handler) def build_hybrid_filtered_map(self, fragfile): """ Mix the fragset with the specified (filtered)filling and then filter the mixture. If the filling is a filtered CSfrag library, this will produce a double-filtered map. @param fragfile: filtered filling (filtered CSfrag fragment file) @type fragfile: str @rtype: L{RosettaFragmentMap} """ if self._matches is None: self.extract_fragments() self.log('\n# Building hybrid filtered map...') filling = [] events = [] def logger(ri): events.append(ri) self._filter_event_handler(ri) try: db = csb.bio.io.wwpdb.FileSystemStructureProvider(self.database) for f in rosetta.RosettaFragmentMap.read(fragfile): filling.append(csb.bio.fragments.Assignment.from_fragment(f, db)) except IOError as io: raise ArgumentIOError(str(io)) except csb.bio.io.wwpdb.StructureNotFoundError as sne: msg = "{0} is not a PDBS25-derived fragset (template {1} not found)" raise ArgumentIOError(msg.format(fragfile, str(sne))) self.log('\n {0} supplementary fragments loaded'.format(len(filling))) self.log('\n Confidence Recurrence Representative Phi Psi Omega', level=2) if len(filling) > self.query.layers.length: msg = "{0} does not look like a filtered fragset (too many centroids)" raise ArgumentError(msg.format(fragfile)) target = csb.bio.fragments.Target.from_profile(self.query) target.assignall(self._matches) target.assignall(filling) factory = csb.bio.fragments.RosettaFragsetFactory() fragset = factory.make_filtered(target, extend=False, callback=logger) return fragset, events class SliceContext(hhsearch.Context): def __init__(self, segment, start, end): self.start = start self.end = end if not isinstance(segment, csb.core.string): segment = segment.to_hmm(convert_scores=True) super(SliceContext, self).__init__(segment) @property def length(self): return self.end - self.start + 1 @property def hits(self): return self.result @property def recurrence(self): return len(self.result) def __lt__(self, other): if self.recurrence == other.recurrence: return self.length < other.length else: return self.recurrence < other.recurrence class PredictionBuilder(object): HEADER = "rank:int residue:str confidence:float centroid:str phi:float psi:float omega:float" @staticmethod def format_angle(angle): """ @param angle: torsion angle value @type angle: float """ if angle is None: return '{0:>6}'.format("-") else: return '{0:6.1f}'.format(angle) @staticmethod def create(ri): """ @param ri: all predictions @type ri: list of L{ResidueEventInfo} """ builder = PredictionBuilder() builder.addall(ri) return builder def __init__(self): self._tsv = csb.io.tsv.Table(PredictionBuilder.HEADER) @property def product(self): """ @rtype: L{Table} """ return self._tsv def add(self, ri): """ @param ri: single residue prediction @type ri: L{ResidueEventInfo} """ row = [ri.rank, repr(ri.type), ri.confidence] if ri.rep: row.append(ri.rep.id) row.append(ri.torsion.phi) row.append(ri.torsion.psi) row.append(ri.torsion.omega) else: row.extend([None] * 4) self.product.insert(row) def addall(self, ri): """ @param ri: all predictions @type ri: list of L{ResidueEventInfo} """ ri = list(ri) ri.sort(key=lambda i: i.rank) for i in ri: self.add(i) def main(): AppRunner().run() if __name__ == '__main__': main()