class HMMModelCreation(ModelCreationAbstractclass):
ex = Exception(
"For the model creation from HMM, input is required. Please select a HMM file using [-i]."
)
open_hmm = lambda _, i: gh.HMMOpen(fileName=i,
filetype=gh.GHMM_FILETYPE_XML)
def create_states(self, **kwargs):
raise NotImplementedError(
"State look-up table creation for HMM currently not available.")
@return_numpy_array
def create_prediction_model(self, **kwargs):
if kwargs["input"] == None:
raise self.ex
hmm = self.open_hmm(kwargs["input"])
trans, _, _ = hmm.asMatrices()
return trans
@return_numpy_array
def create_observation_model(self, **kwargs):
if kwargs["input"] == None:
raise self.ex
hmm = self.open_hmm(kwargs["input"])
_, emi, _ = hmm.asMatrices()
return emi
def tffm_from_xml(xml, kind):
"""
Construct a TFFM described in an XML file.
:arg xml: File containing the TFFM description in XML format.
:type xml: str
:arg kind: Type of TFFM to construct between '1st-order' and 'detailed'.
:type kind: str
:returns: The TFFM described in the XML file.
:rtype: :class:`TFFM`
"""
hmm = ghmm.HMMOpen(xml)
return TFFM(hmm.emissionDomain, hmm.distribution, hmm.cmodel, kind)
def __create_hmm_from_xml(self, xml):
"""Creates a hmm from the xml representation. Not nice to use tempfile
but not otherwise possible due to hidden code and swig in ghmm.
:param xml: The xml string
:return: the ghmm hmm object
"""
fd, name = tempfile.mkstemp()
f = os.fdopen(fd, 'w')
f.write(xml)
f.close(
) # File has to be closed before reading from it. Otherwise this will fail.
hmm = gh.HMMOpen(fileName=name, filetype=gh.GHMM_FILETYPE_XML)
os.remove(name)
return hmm
def __init__(self, allele):
''' Initialize the predictor. For now, only H2-IAb is supported. '''
if (allele != 'H2-IAb'):
raise ValueError('Unsupported allele')
self.allele = allele
# range of allowed peptide lengths (percentile rank calibration was
# done for this range)
self.minPeptideLength = 12
self.maxPeptideLength = 24
# define the HMM emission alphabet, including a dedicated "stop"
# emission (a GHMM-specific tweak)
self.alphabet = ghmm.Alphabet(ghmm.AminoAcids.listOfCharacters + ['Z'])
# a value to replace inf in the predictions. Same value as used in
# percentile rank calibration
self.infSubstitute = 1000
# GHMM has an internal limitation of 1,500,000 sequences in a set
self.sequencesPerBlock = 1500000
# load the HMM model and the corresponding percentile rank calibration data
from pkg_resources import resource_filename
hmmFile = resource_filename(__name__,
'models/hmm-h2iab-iedb2018-binders.xml')
prCalibrationFile = resource_filename(
__name__, 'models/precent-rank-model-h2iab.npz')
self.hmm = ghmm.HMMOpen(hmmFile)
prCalibration = np.load(prCalibrationFile)
self.prCalibrationCdf = prCalibration['cdf']
self.prCalibrationBinEdges = prCalibration['bin_edges']
#!/usr/bin/env python
# -*- coding:utf-8 -*-
import ghmm
import hmm
import predictor
import pickle
# make a dataset
dm = predictor.FastaDataSetMaker()
ta = dm.read_from_file("TA_all.fasta", name="ta", label=1)
charlist = "ACDEFGHIKLMNPQRSTVWY"
char_dic = {charlist[i]: i for i in xrange(20)}
ta_n = [[char_dic[c] for c in seq.sequence[::-1]] for seq in ta]
# convert hmm
g = ghmm.HMMOpen("ta2.1.2.xml")
h = hmm.convert_ghmm(g)
# Training
h.baum_welch(ta_n, iter_limit=1000, threshold=-1e-3, pseudocounts=[0, 1e-3, 0])
g2 = hmm.convert2ghmm(h)
f = open("ta_with_noise", "w")
pickle.dump(g2, f)
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import argparse
import ghmm as gh
import json
from qsrrep_lib.rep_hmm import RepHMM
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-i", "--input", help="The xml HMM file to convert", type=str)
parser.add_argument("-o", "--output", help="The new filename", type=str)
args = parser.parse_args()
hmm = gh.HMMOpen(fileName=args.input, filetype=gh.GHMM_FILETYPE_XML)
trans, emi, start = hmm.asMatrices()
ret = {
RepHMM.TRANS: trans,
RepHMM.EMI: emi,
RepHMM.START: start,
}
with open(args.output, 'w') as f:
json.dump(ret, f)