Implement subsampling via a script

This implements a new script to encapsulate the subsampling logic formerly encoded in snakemake rules. This is in preparation for moving this script to the augur repo where it will become `augur subsample`. (We have chosen to develop this in the ncov repo for simplicity.)

The script currently uses the same approach as the former snakemake rules, however python functions are called rather than scripts / augur commands. Briefly, the steps are:

1. A subsampling scheme is provided, parsed, validated, and turned into a simple graph to indicate which samples rely on other samples having been computed (i.e. which are needed for priorities)
2. Each sample is computed by calling the run function of augur filter
3. If priorities need to be calculated for a sample to be computed, this is achieved by calling functions from the two existing scripts.
4. The set of sequences to include in each sample is combined, and outputs written.

Author: jameshadfield

scripts/subsample.py

@@ -0,0 +1,319 @@
+from augur.utils import AugurException
+from augur.filter import run as augur_filter
+from augur.index import index_sequences
+from augur.io import write_sequences, open_file, read_sequences, read_metadata
+from get_distance_to_focal_set import get_distance_to_focal_set # eventually from augur.priorities (or similar)
+from priorities import create_priorities # eventually from augur.priorities (or similar)
+import yaml
+from argparse import Namespace, ArgumentParser, ArgumentDefaultsHelpFormatter
+from os import path
+import pandas as pd
+from tempfile import NamedTemporaryFile
+import jsonschema
+# from pkg_resources import resource_string
+
+DESCRIPTION = "Subsample sequences based on user-defined YAML configuration"
+
+def register_arguments(parser):
+    parser.add_argument('--scheme', required=True, metavar="YAML", help="subsampling scheme")
+    parser.add_argument('--output-dir', required=True, metavar="PATH", help="directory to save intermediate results")
+    parser.add_argument('--metadata', required=True, metavar="TSV", help="metadata")
+    parser.add_argument('--alignment', required=True, metavar="FASTA", help="alignment to subsample")
+    parser.add_argument('--alignment-index', required=False, metavar="INDEX", help="sequence index of alignment")
+    parser.add_argument('--reference', required=True, metavar="FASTA", help="reference (which was used for alignment)")
+    parser.add_argument('--include-strains-file', required=False, nargs="+", default=None, metavar="TXT", help="strains to force include")
+    parser.add_argument('--exclude-strains-file', required=False, nargs="+", default=None, metavar="TXT", help="strains to force exclude")
+    parser.add_argument('--output-fasta', required=True, metavar="FASTA", help="output subsampled sequences")
+    parser.add_argument('--output-metadata', required=True, metavar="TSV", help="output subsampled metadata")
+    parser.add_argument('--output-log', required=False, metavar="TSV", help="log file explaining why strains were excluded / included")
+
+def run(args):
+
+    config = parse_scheme(args.scheme)
+
+    generate_sequence_index(args)
+
+    samples = [Sample(name, data, args) for name, data in config.items()]
+
+    graph = make_graph(samples)
+
+    traverse_graph(
+        graph,
+        lambda s: s.filter()
+    )
+
+    combine_samples(args, samples)
+
+def parse_scheme(filename):
+    with open(filename) as fh:
+        try:
+            data = yaml.safe_load(fh)
+        except yaml.YAMLError as exc:
+            print(exc)
+            raise AugurException(f"Error parsing subsampling scheme {filename}")
+    validate_scheme(data)
+    return data
+
+
+def validate_scheme(scheme):
+    try:
+        # When we move this to `augur subsample`, load the schema via:
+        # schema = yaml.safe_load(resource_string(__package__, path.join("data", "schema-subsampling.yaml")))
+        with open(path.join(path.dirname(path.realpath(__file__)), "subsample_schema.yaml")) as fh:
+            schema = yaml.safe_load(fh)
+    except yaml.YAMLError as err:
+        raise AugurException("Subsampling schema definition is not valid YAML. Error: {}".format(err))
+    # check loaded schema is itself valid -- see http://python-jsonschema.readthedocs.io/en/latest/errors/
+    try:
+        jsonschema.Draft6Validator.check_schema(schema)
+    except jsonschema.exceptions.SchemaError as err:
+        raise AugurException("Subsampling schema definition is not valid. Error: {}".format(path, err))
+
+    try:
+        jsonschema.Draft6Validator(schema).validate(scheme)
+    except jsonschema.exceptions.ValidationError as err:
+        print(err)
+        raise AugurException("Subsampling scheme failed validation")
+
+class Sample():
+    """
+    A class to hold information about a sample. A subsampling scheme will consist of multiple
+    samples. Each sample may depend on the priorities based off another sample.
+    """
+    def __init__(self, name, config, cmd_args):
+        self.name = name
+        self.tmp_dir = cmd_args.output_dir
+        self.alignment = cmd_args.alignment
+        self.alignment_index = cmd_args.alignment_index
+        self.reference = cmd_args.reference
+        self.metadata = cmd_args.metadata
+        self.initialise_filter_args(config, cmd_args)
+        self.priorities = config.get("priorities", None) 
+        print("Constructor", self.name)
+
+    def initialise_filter_args(self, config, subsample_args):
+        """
+        Currently this method is needed as we need to call `augur filter`'s `run()` with an
+        argparse instance. An improvement here would be to expose appropriate filtering
+        functions and call them as needed, with the output being returned rather than
+        written to disk.
+        """
+        args = Namespace()
+        args.metadata = self.metadata
+        args.sequences = self.alignment
+        args.sequence_index = self.alignment_index
+        args.metadata_chunk_size = 100000
+        args.metadata_id_columns = ["strain", "name"]
+        args.min_date = None
+        args.max_date = None
+        args.group_by = config.get('group-by', None)
+        args.sequences_per_group = config.get("sequences-per-group", None)
+        args.subsample_max_sequences = config.get("subsample-max-sequences", None)
+        args.exclude_ambiguous_dates_by = None
+        args.exclude = subsample_args.exclude_strains_file
+        args.exclude_all = None
+        args.exclude_where = config.get('exclude-where', None)
+        args.include = subsample_args.include_strains_file
+        args.include_where = None
+        args.min_length = None
+        args.non_nucleotide = None
+        args.probabilistic_sampling = config.get("probabilistic-sampling", None)
+        args.no_probabilistic_sampling = config.get("no-probabilistic-sampling", None)
+        args.priority = None
+        args.subsample_seed = None
+        args.query = config.get("query", None)
+        args.output          = path.join(self.tmp_dir, f"sample.{self.name}.fasta") # filtered sequences in FASTA forma
+        args.output_metadata = path.join(self.tmp_dir, f"sample.{self.name}.tsv")   # metadata for strains that passed filters
+        args.output_strains  = path.join(self.tmp_dir, f"sample.{self.name}.txt")   # list of strains that passed filters (no header)
+        args.output_log      = path.join(self.tmp_dir, f"sample.{self.name}.log.tsv")
+        self.filter_args = args
+
+
+    def calculate_required_priorities(self):
+        """
+        If computation of this sample requires priority information of another sample
+        (the "focus"), then this function will compute those priorities.
+        """
+        if not self.priorities:
+            return
+        focal_sample = self.priorities.get('sample', None)
+        if not focal_sample:
+            raise AugurException(f"Cannot calculate priorities needed for {self.name} as the {self.get_priority_focus_name()} sample wasn't linked")
+        print(f"Calculating priorities required by {self.name}")
+        priorities_file = focal_sample.calculate_priorities()
+        self.filter_args.priority = priorities_file
+
+    def calculate_priorities(self):
+        """
+        Calculate the priorities TSV file for the alignment in the context of this sample
+
+        Returns the filename of the priorities file (TSV)
+        """
+        print(f"Calculating proximity of {self.name}")
+        get_distance_to_focal_set(
+            self.alignment,
+            self.reference,
+            self.filter_args.output,
+            path.join(self.tmp_dir, f"proximity_{self.name}.tsv"),
+            ignore_seqs=["Wuhan/Hu-1/2019"]  # TODO - use the config to define this?
+        )
+        print(f"Calculating priorities of {self.name}")
+        priorities_path = path.join(self.tmp_dir, f"priorities_{self.name}.tsv")
+        create_priorities(
+            self.alignment_index,
+            path.join(self.tmp_dir, f"proximity_{self.name}.tsv"),
+            priorities_path
+        )
+        return priorities_path
+
+    def get_priority_focus_name(self):
+        if not self.priorities:
+            return None
+        return self.priorities['focus']
+
+    def set_priority_sample(self, sample):
+        if not self.priorities:
+            raise AugurException(f"No priorities set for {self.name}")
+        self.priorities['sample'] = sample
+
+    def filter(self):
+        print("\n---------------------------------\nCONSTRUCTING SAMPLE FOR", self.name, "\n---------------------------------")
+        self.calculate_required_priorities()
+        augur_filter(self.filter_args)
+        # In the future, instead of `augur_filter` saving data to disk, it would return
+        # data to the calling process. In lieu of that, we read the data just written.
+        self.sampled_strains = set(pd.read_csv(self.filter_args.output_strains, header=None)[0])
+        self.filter_log = pd.read_csv(
+            self.filter_args.output_log,
+            header=0,
+            sep="\t",
+            index_col="strain"
+        )
+
+
+def make_graph(samples):
+    """"
+    Given a config file, construct a graph of samples to perform in an iterative fashion, such that
+    priorities 
+    This is a DAG, however an extremely simple one which we can construct outselves rather than relying on
+    extra libraries.
+    Constraints:
+    * Each sample can only use priorities of one other sample
+    * Acyclic
+    Structure:
+    tuple: (sample name, list of descendent samples) where a "descendant" sample requires the linked sample to be
+    created prior to it's creation. Each entry in the list has this tuple structure.
+    """
+
+    included = set() # set of samples added to graph
+    graph = (None, [])
+
+    # add all the samples which don't require priorities to the graph
+    for sample in samples:
+        if not sample.get_priority_focus_name():
+            graph[1].append((sample, []))
+            included.add(sample.name)
+
+    def add_descendants(level):
+        parent_sample = level[0]
+        descendants = level[1]
+        for sample in samples:
+            if sample.name in included:
+                continue
+            if sample.get_priority_focus_name() == parent_sample.name:
+                sample.set_priority_sample(parent_sample)
+                descendants.append((sample, []))
+                included.add(sample.name)
+        for inner_level in descendants:
+            add_descendants(inner_level)
+
+    for level in graph[1]:
+        add_descendants(level)
+
+    # from pprint import pprint
+    # print("\ngraph"); pprint(graph);print("\n")
+
+    if len(samples)!=len(included):
+        AugurException("Incomplete graph construction")
+
+    return graph
+
+def traverse_graph(level, callback):
+    this_sample, descendents = level
+    if this_sample:
+        callback(this_sample)
+    for child in descendents:
+        traverse_graph(child, callback)
+
+def generate_sequence_index(args):
+    if args.alignment_index:
+        print("Skipping sequence index creation as an index was provided")
+        return    
+    print("Creating ephemeral sequence index file")
+    with NamedTemporaryFile(delete=False) as sequence_index_file:
+        sequence_index_path = sequence_index_file.name
+        index_sequences(args.alignment, sequence_index_path)
+        args.alignment_index = sequence_index_path
+
+
+def combine_samples(args, samples):
+    """Collect the union of strains which are included in each sample and write them to disk.
+    Parameters
+    ----------
+    args : argparse.Namespace
+        Parsed arguments from argparse
+    samples : list[Sample]
+        list of samples
+    """
+    print("\n\n")
+    ### Form a union of each sample set, which is the subsampled strains list
+    sampled_strains = set()
+    for sample in samples:
+        print(f"Sample \"{sample.name}\" included {len(sample.sampled_strains)} strains")
+        sampled_strains.update(sample.sampled_strains)
+    print(f"In total, {len(sampled_strains)} strains are included in the resulting subsampled dataset")
+
+    ## Iterate through the input sequences, streaming a subsampled version to disk.
+    sequences = read_sequences(args.alignment)
+    sequences_written_to_disk = 0
+    with open_file(args.output_fasta, "wt") as output_handle:
+        for sequence in sequences:
+            if sequence.id in sampled_strains:
+                sequences_written_to_disk += 1
+                write_sequences(sequence, output_handle, 'fasta')
+    print(f"Writing {sequences_written_to_disk} sequences to {args.output_fasta}")
+
+    ## Iterate through the metadata in chunks, writing out those entries which are in the subsample
+    metadata_reader = read_metadata(
+        args.metadata,
+        id_columns=["strain", "name"], # TODO - this should be an argument
+        chunk_size=10000 # TODO - argument
+    )
+    metadata_header = True
+    metadata_mode = "w"
+    metadata_written_to_disk = "???" # TODO
+    for metadata in metadata_reader:
+        metadata.loc[sampled_strains].to_csv(
+            args.output_metadata,
+            sep="\t",
+            header=metadata_header,
+            mode=metadata_mode,
+        )
+        metadata_header = False
+        metadata_mode = "a"
+    print(f"Writing {metadata_written_to_disk} metadata entries sequences to {args.output_metadata}")
+
+    ## Combine the log files (from augur filter) for each sample into a larger log file
+    ## Format TBD
+    ## TODO
+
+if __name__ == "__main__":
+    # the format of this block is designed specifically for future transfer of this script
+    # into augur in the form of `augur subsample`
+    parser = ArgumentParser(
+        usage=DESCRIPTION,
+        formatter_class=ArgumentDefaultsHelpFormatter,
+    )
+    register_arguments(parser)
+    args = parser.parse_args()
+    run(args)

scripts/subsample_schema.yaml

@@ -0,0 +1,49 @@
+
+type: object
+title: YAML Schema for subsampling configuration to be consumed by a subsample script / command
+patternProperties:
+    "^[a-zA-Z0-9*_-]+$":
+        type: object
+        title: description of a sample
+        additionalProperties: false
+        properties:
+            group-by:
+                type: array
+                minItems: 1
+                items:
+                    type: string
+            sequences-per-group:
+                type: integer
+            subsample-max-sequences:
+                type: integer
+            exclude_ambiguous_dates_by:
+                type: string
+                enum: ["any", "day", "month", "year"]
+            query:
+                type: string
+            min-date:
+                type: ["number", "string"]
+                pattern: ^\d{4}-\d{2}-\d{2}$
+            max-date:
+                type: ["number", "string"]
+                pattern: ^\d{4}-\d{2}-\d{2}$
+            exclude-where:
+                type: array
+                minItems: 1
+                items:
+                    type: string
+            include-where:
+                type: array
+                minItems: 1
+                items:
+                    type: string
+            query:
+                type: string
+            probabilistic-sampling:
+                type: boolean
+            no-probabilistic-sampling:
+                type: boolean
+            no_subsampling:
+                type: boolean
+            priorities:
+                type: object