def main():
'''
Demo to create a feature vector for protein fold classification.
In this demo we try to classify a protein chain as either an
all alpha or all beta protein based on protein sequence. We use
n-grams and Word2Vec representation of the protein sequence as a
feature vector.
'''
start = time.time()
conf = SparkConf() \
.setMaster("local[*]") \
.setAppName("featuresDemo")
sc = SparkContext(conf=conf)
path = "../../resources/mmtf_reduced_sample/"
# Read MMTF Hadoop sequence file and create a non-redundant set (<=40% seq. identity)
# of L-protein chains
sequenceIdentity = 40
resolution = 2.0
pdb = MmtfReader \
.readSequenceFile(path, sc) \
.filter(pisces(sequenceIdentity, resolution)) \
.flatMap(structureToPolymerChains()) \
.filter(pisces(sequenceIdentity, resolution)) \
.filter(containsLProteinChain()) \
# Get secondary structure content
data = secondaryStructureExtractor.getDataset(pdb)
# classify chains by secondary structure type
minThreshold = 0.05
maxThreshold = 0.15
data = addProteinFoldType(data, minThreshold, maxThreshold)
# add Word2Vec encoded feature vector
encoder = proteinSequenceEncoder(data)
n = 2 # Create 2-grams
windowSize = 25 # 25-amino residue window size for Word2Vec
vectorSize = 50 # dimension of feature vector
data = encoder.overlappingNgramWord2VecEncode(n, windowSize,
vectorSize).cache()
data.printSchema()
data.show(25)
# keep only a subset of relevant fields for futher processing
data = data.select("structureChainId", "alpha", "beta", "coil", "foldType",
"features")
data.write.mode("overwrite").format("parquet").save(
"/home/marshuang80/PDB/data/demo.parquet")
end = time.time()
print("Time: %f sec." % (end - start))
sc.stop()
def main():
'''
This class creates a dataset of sequence segments dericed from a
non-redundant set. The dataset contains the sequence segment, the DSSP Q8
and DSSP Q3 code of the center residue in a seuqnce segment, and a one-hot
encoding of the sequence segment.
The dataset is saved in a file specified by the user
'''
start = time.time()
conf = SparkConf() \
.setMaster("local[*]") \
.setAppName("SecondaryStructureOneHotEncoderDemo")
sc = SparkContext(conf=conf)
# Read MMTF Hadoop sequence file and create a non-redundant set
# (<=20% seq. identity) of L-protein chains
path = "../../resources/mmtf_reduced_sample/"
sequenceIdentity = 20
resolution = 2.0
fraction = 0.1
seed = 123
#.downloadMmtfFiles(["2ONX",'1JLP','5X6H','5L2G','2MK1' ],sc) \
pdb = MmtfReader \
.readSequenceFile(path, sc) \
.flatMap(structureToPolymerChains()) \
.filter(pisces(sequenceIdentity, resolution)) \
.filter(containsLProteinChain()) \
.sample(False, fraction, seed)
segmentLength = 11
data = secondaryStructureSegmentExtractor.getDataset(
pdb, segmentLength).cache()
print(f"original data : {data.count()}")
data = data.dropDuplicates(["labelQ3", "sequence"]).cache()
print(f"- duplicate Q3/seq : {data.count()}")
data = data.dropDuplicates(["sequence"])
print(f"- duplicate seq : {data.count()}")
encoder = proteinSequenceEncoder(data)
data = encoder.oneHotEncode()
data.printSchema()
data.show(25, False)
end = time.time()
print("Time: %f sec." % (end - start))
sc.stop()
def main():
'''
This class creates a dataset of sequence segment derived from a
non-redundant set. The dataset contains the sequence segment, the DSSP
Q8 and DSSP Q3 code of the center residue in a sequence segment, and a
Word2Vec encoding of the sequence segment.
The Data is saved in JSON file specified by the user.
'''
start = time.time()
conf = SparkConf() \
.setMaster("local[*]") \
.setAppName("secondaryStructureWord2VecEncodeDemo")
sc = SparkContext(conf=conf)
# Read MMTF Hadoop sequence file and create a non-redundant set
# (<=20% seq. identity) of L-protein chains
path = "../../resources/mmtf_reduced_sample/"
sequenceIdentity = 20
resolution = 2.0
fraction = 0.1
seed = 123
pdb = MmtfReader \
.readSequenceFile(path, sc) \
.flatMap(structureToPolymerChains()) \
.filter(pisces(sequenceIdentity, resolution)) \
.filter(containsLProteinChain()) \
.sample(False, fraction, seed)
segmentLength = 11
data = secondaryStructureSegmentExtractor.getDataset(
pdb, segmentLength).cache()
# add Word2Vec encoded feature vector
encoder = proteinSequenceEncoder(data)
n = 2
windowSize = (segmentLength - 1) // 2
vectorSize = 50
data = encoder.overlappingNgramWord2VecEncode(n, windowSize, vectorSize)
data.printSchema()
data.show(25, False)
end = time.time()
print("Time: %f sec." % (end - start))
sc.stop()