genotypes.py

"""Expose API to Genotypes and VCFs."""
from collections import OrderedDict
import copy
import json
from flask import request
import flask.ext.restful as restful
from sqlalchemy import (select, func, types, cast, join, outerjoin, asc, desc,
                        and_, Integer, Float, String, distinct)
from sqlalchemy.sql import text
from sqlalchemy.sql.expression import label, column, case, literal
from sqlalchemy.sql.functions import coalesce
import vcf as pyvcf
from plone.memoize import forever

from cycledash import db
from common.helpers import tables

from . import Resource


class Genotypes(Resource):
    require_auth = True
    def get(self, run_id):
        return get(run_id, json.loads(request.args.get('q')))


  ##############################################################################
 ##### The below functions are exposed via the controllers in views.py. #######
##############################################################################

@forever.memoize
def spec(vcf_id):
    """Return a dict describing the layout of the columns to be displayed.

    In form:
    {INFO: {attr1: {path: ['INFO', 'attr1'],
                          name: 'attr1',
                          info: {type: 'Integer', description: 'This col. et'}},
            ...},
    SAMPLE: {attrA: ...},
    ANNOTATIONS: {attrA: ...}}
    """
    with tables(db.engine, 'vcfs') as (con, vcfs):
        q = (select([vcfs.c.vcf_header, vcfs.c.extant_columns])
                    .where(vcfs.c.id == vcf_id))
        res = con.execute(q).fetchone()
    return _header_spec(res['vcf_header'], res['extant_columns'])


@forever.memoize
def samples(vcf_id):
    """Return a sorted list of sample names found in the given vcf."""
    query = """SELECT sample_name FROM genotypes WHERE vcf_id = %s
    GROUP BY sample_name ORDER BY sample_name
    """
    with tables(db.engine, 'genotypes') as (con, genotypes):
        samples = (select([func.count(distinct(genotypes.c.sample_name))])
                   .where(genotypes.c.vcf_id == vcf_id))
        samples = [sample_name for (sample_name,)
                   in samples.execute().fetchall()]
    return samples


@forever.memoize
def contigs(vcf_id):
    """Return a sorted list of contig names found in the given vcf."""
    with tables(db.engine, 'genotypes') as (con, genotypes):
        q = (select([genotypes.c.contig])
             .where(genotypes.c.vcf_id == vcf_id)
             .group_by(genotypes.c.contig)
             .order_by(func.length(genotypes.c.contig), genotypes.c.contig))
        results = con.execute(q).fetchall()
    return [contig for (contig,) in results]


def get(run_id, query, with_stats=True):
    """Return a list of genotypes in a vcf conforming to the given query, as
    well as a dict of stats calculated on them.

    If a truth_vcf is associated with this VCF, stats include true/false,
    positive/negative stats, as well as precision, recall, and f1score. Stats
    also include the number of records, and the number of records once filters
    are applied.

    A query is a dictionary which specifies the range, filters, limit, offset
    and ordering which should be applied against genotypes before genotypes and
    stats are returned.

    It has structure:

    {range: {contig: "X", start: 0, end: 250000000},
     filters: [{columnName: 'info:DP', filterValue: '50', type: '<'}, ...],
     sortBy: [{columnName: 'contig', order: 'asc'},
              {columnName: 'position', order: 'asc'}, ...],
     page: 10,
     limit: 250
    }
    """
    query = _annotate_query_with_types(query, spec(run_id))
    compare_to_run_id = query.get('compareToVcfId')
    with tables(db.engine, 'genotypes') as (con, g):
        if compare_to_run_id:
            # We consider a genotype validated if a truth genotype exists at its
            # location (contig/position) with the same ref/alts.  This isn't
            # entirely accurate: for example, it handles SVs very poorly.
            gt = g.alias()
            joined_q = outerjoin(g, gt, and_(
                gt.c.vcf_id == compare_to_run_id,
                g.c.contig == gt.c.contig,
                g.c.position == gt.c.position,
                g.c.reference == gt.c.reference,
                g.c.alternates == gt.c.alternates,
                g.c.sample_name == gt.c.sample_name))
            valid_column = label('tag:true-positive', gt.c.contig != None)
            q = (select(g.c + [valid_column])
                 .select_from(joined_q)
                 .where(g.c.vcf_id == run_id))
        else:
            q = select(g.c).where(g.c.vcf_id == run_id)

        q = _add_range(q, g, query.get('range'))
        q = _add_filters(q, g, query.get('filters'))
        q = _add_orderings(q, g, query.get('sortBy'))
        q = _add_paging(q, g, query.get('limit'), query.get('page'))

        q = _add_ordering(q, g, 'String', 'contig', 'asc')
        q = _add_ordering(q, g, 'Integer', 'position', 'asc')
        genotypes = [dict(g) for g in con.execute(q).fetchall()]
    stats = calculate_stats(run_id, compare_to_run_id, query) if with_stats else {}
    return {'records': genotypes, 'stats': stats}


@forever.memoize
def calculate_stats(vcf_id, truth_vcf_id, query):
    """Return stats for genotypes in vcf and truth_vcf conforming to query.

    Args:
       vcf_id: the vcf being examined.
       truth_vcf_id: the truth_vcf being validated against.
       query: the query object.
    """
    with tables(db.engine, 'genotypes') as (con, genotypes):
        # The number of records being displayed:
        count_q = select([func.count()]).where(genotypes.c.vcf_id == vcf_id)
        count_q = _add_filters(count_q, genotypes, query.get('filters'))
        count_q = _add_range(count_q, genotypes, query.get('range'))
        (count,) = con.execute(count_q).fetchone()

        # The total number of records in VCF:
        total_count_q = select([func.count()]).where(
            genotypes.c.vcf_id == vcf_id)
        (total_count,) = con.execute(total_count_q).fetchone()
    return genotype_statistics(query, vcf_id, truth_vcf_id, count, total_count)


def genotype_statistics(query, vcf_id, truth_vcf_id, count, total_count):
    """Return precision, recall, f1score, and count statistics for the given
    query on vcf, validated against truth_vcf.
    """
    stats = {'totalRecords': count, 'totalUnfilteredRecords': total_count}

    if not truth_vcf_id:
        return stats

    with tables(db.engine, 'genotypes') as (con, genotypes):
        g, gt = genotypes.alias(), genotypes.alias()
        joined_q = join(g, gt, and_(g.c.contig == gt.c.contig,
                                    g.c.position == gt.c.position,
                                    g.c.reference == gt.c.reference,
                                    g.c.alternates == gt.c.alternates,
                                    g.c.sample_name == gt.c.sample_name))
        true_pos_q = select([func.count()]).select_from(joined_q).where(
            and_(g.c.vcf_id == vcf_id, gt.c.vcf_id == truth_vcf_id))
        true_pos_q = _add_filters(true_pos_q, g, query.get('filters'))
        true_pos_q = _add_range(true_pos_q, g, query.get('range'))

        query = _whitelist_query_filters(query)
        true_pos_q = _add_filters(true_pos_q, gt, query.get('filters'))
        true_pos_q = _add_range(true_pos_q, gt, query.get('range'))
        (true_positives,) = con.execute(true_pos_q).fetchone()

        # This calculates the total number of truth records given a subset of
        # the filters which makes sense to apply to a validation set.
        total_truth_q = select(
            [func.count()]).select_from(g).where(g.c.vcf_id == truth_vcf_id)
        total_truth_q = _add_filters(total_truth_q, g, query.get('filters'))
        total_truth_q = _add_range(total_truth_q, g, query.get('range'))
        (total_truth,) = con.execute(total_truth_q).fetchone()
        total_truth *= len(samples(truth_vcf_id))

    stats.update(_true_false_pos_neg(true_positives, total_truth, count))
    return stats


def genotypes_for_records(vcf_id, query):
    """Return all genotypes which would appear on a row in a VCF (determined by
    CHROM/POS/REF/ALT) if just one genotype on that row passes the selections in
    `query'.

    This is used to generate the list of genotypes to be transformed into
    vcf.model._Records and then written to a VCF file.
    """
    query = _annotate_query_with_types(query, spec(vcf_id))
    with tables(db.engine, 'genotypes') as (con, gt):
        keyfunc = func.concat(
            gt.c.contig, ':', cast(gt.c.position, types.Unicode), '::',
            gt.c.reference, '->', gt.c.alternates)
        filtered_gts_q = select([keyfunc]).where(gt.c.vcf_id == vcf_id)
        filtered_gts_q = _add_filters(filtered_gts_q, gt, query.get('filters'))
        filtered_gts_q = _add_range(filtered_gts_q, gt, query.get('range'))
        filtered_gts_q = filtered_gts_q.cte('filtered_gts')

        records_q = select([gt]).where(
            keyfunc.in_(select([filtered_gts_q]))).where(gt.c.vcf_id == vcf_id)
        records_q = records_q.order_by(asc(func.length(gt.c.contig)),
                                       asc(gt.c.contig),
                                       asc(gt.c.position),
                                       asc(gt.c.reference),
                                       asc(gt.c.alternates),
                                       asc(gt.c.sample_name))
        genotypes = [dict(g) for g in con.execute(records_q).fetchall()]
    return genotypes


   ############################################################################
  ### The below functions are helpers used to generate a SQLAlchemy select  ##
 ## object.                                                                ##
############################################################################

def _add_orderings(sql_query, table, sort_by):
    """
    Args:
      sort_by: a list of
        {order: asc | desc, columnName: name, sqlTyle: float | integer}
    """
    for sort_spec in sort_by:
        column_name = sort_spec['columnName']
        column_type = sort_spec.get('columnType', 'integer')
        order = sort_spec.get('order', 'asc')
        sql_query = _add_ordering(sql_query, table,
                                  column_type, column_name, order)
    return sql_query


def _add_ordering(sql_query, table, column_type, column_name, order):
    # Special case for this column, which sorts contigs correctly:
    if column_name == 'contig':
        get_contig_num = cast(
            text("SUBSTRING({} FROM '\d+')".format(table.c.contig)),
            type_=Integer)
        starts_with_chr = (text("SUBSTRING({} FROM '^chr(\d+)')"
                                .format(table.c.contig)) != literal(''))
        starts_with_number = (text("SUBSTRING({} FROM '^\d+')"
                                   .format(table.c.contig)) != literal(''))
        # 10000 used here to mean "should be at the end of all the numbers",
        # assuming we never hit a chromosome number >= 10000.
        contig_num_col = case(
            [(starts_with_chr, get_contig_num),
             (starts_with_number, get_contig_num)],
            else_=literal(10000)
        )
        contig_len_col = func.length(table.c.contig)
        contig_col = table.c.contig
        if order == 'desc':
            contig_len_col = desc(contig_len_col)
            contig_col = desc(contig_col)
        return sql_query.order_by(contig_num_col, contig_len_col, contig_col)
    sqla_type = vcf_type_to_sqla_type(column_type)
    column = cast(table.c[column_name], type_=sqla_type)
    column = {'asc': asc(column), 'desc': desc(column)}.get(order)
    return sql_query.order_by(column)


def _add_paging(sql_query, table, limit, page):
    if limit:
        limit = int(limit)
        page = int(page)
        offset = limit * page
        sql_query = sql_query.limit(limit).offset(offset)
    return sql_query


def _add_filters(sql_query, table, filters):
    for filt in filters:
        column_name = filt.get('columnName')
        column_type = filt.get('columnType')
        value = filt.get('filterValue')
        op_name = filt.get('type')
        sql_query = _add_filter(
            sql_query, table, column_name, column_type, value, op_name)
    return sql_query


def _add_filter(sql_query, table, column_name, column_type, value, op_name):
    sqla_type = vcf_type_to_sqla_type(column_type)
    col = column(column_name)
    if table.c.get(column_name) is not None:
        col = cast(table.c[column_name], sqla_type)
    if op_name == '=':    return sql_query.where(col == value)
    elif op_name == '!=': return sql_query.where(col != value)
    elif op_name == '<':  return sql_query.where(col < value)
    elif op_name == '>':  return sql_query.where(col > value)
    elif op_name == '>=': return sql_query.where(col >= value)
    elif op_name == '<=': return sql_query.where(col <= value)
    elif op_name == 'NULL': return sql_query.where(col == None)
    elif op_name == 'NOT NULL': return sql_query.where(col != None)
    elif op_name == 'LIKE': return sql_query.where(col.like(value))
    elif op_name == 'RLIKE': return sql_query.where(col.op('~*')(value))


def _add_range(sql_query, table, rangeq):
    rangeq = rangeq or {}
    contig = rangeq.get('contig')
    start = rangeq.get('start')
    end = rangeq.get('end')
    if contig:
        sql_query = sql_query.where(table.c.contig == contig)
    if start:
        sql_query = sql_query.where(table.c.position >= start)
    if end:
        sql_query = sql_query.where(table.c.position < end)
    return sql_query


def vcf_type_to_sqla_type(column_type='integer'):
    column_type = column_type.lower()
    return {'integer': Integer, 'float': Float}.get(column_type, String)


  ######################
 ### Helper methods: ##
######################

def _true_false_pos_neg(true_positives, total_truth_records, total_records):
    false_negatives = total_truth_records - true_positives
    false_positives = total_records - true_positives

    def safe_div(numerator, denom):
        return float(numerator) / denom if denom > 0 else 0

    precision = safe_div(true_positives, true_positives + false_positives)
    recall = safe_div(true_positives, true_positives + false_negatives)
    f1score = safe_div(2 * (precision * recall), precision + recall)

    return {
        'truePositives': true_positives,
        'falsePositives': false_positives,
        'falseNegatives': false_negatives,
        'precision': precision,
        'recall': recall,
        'f1score': f1score,
        'totalTruthRecords': total_truth_records
    }


def _header_spec(vcf_header_text, extant_cols):
    """Return dict representation of a CycleDash header."""
    reader = pyvcf.Reader(line for line in vcf_header_text.split('\n'))
    res = OrderedDict()
    for (supercolumn, attr) in [('info', 'infos'), ('sample', 'formats')]:
        res[supercolumn.upper()] = OrderedDict()
        for key, val in reader.__dict__[attr].iteritems():
            column_name = supercolumn + ':' + val.id
            if column_name not in extant_cols:
                continue
            _add_column_to_spec(
                spec=res,
                column_name=column_name,
                supercolumn=supercolumn,
                subcolumn=key,
                column_type=val.type,
                num=val.num,
                description=val.desc)

    # Sample name is not a part of the SAMPLE: hierarchy, but we want to add it
    # into that hierarchy for display purposes.
    _add_column_to_spec(
        spec=res,
        column_name='sample_name',
        supercolumn='SAMPLE',
        subcolumn='Sample Name',
        column_type='String',
        num=1,
        description='The name of the sample',
        path=['sample_name']) # This path is not the default super -> sub column

    # Add Cycledash-derived columns
    _add_extant_column_to_spec(extant_cols, 'annotations:gene_name', res,
            ('Name of the gene that overlaps with this variant\'s effect'
            'derived from Varcode project.'))
    _add_extant_column_to_spec(extant_cols, 'annotations:transcript', res,
            ('Transcript that overlaps with this variant'
            'derived from Varcode project.'))
    _add_extant_column_to_spec(extant_cols, 'annotations:effect_notation', res,
            ('Protein change caused by this variant, '
            'derived from Varcode project.'))
    _add_extant_column_to_spec(extant_cols, 'annotations:effect_type', res,
            ('Type of the variant, '
            'derived from Varcode project.'))

    # Remove empty supercolumns
    for key, val in res.iteritems():
        if not val.keys():
            del res[key]

    return res

def _add_extant_column_to_spec(extant_cols, column_name, res, description):
   if column_name in extant_cols:
        supercolumn, subcolumn = column_name.split(':')
        _add_column_to_spec(
            spec=res,
            column_name=column_name,
            supercolumn=supercolumn,
            subcolumn=subcolumn,
            column_type='String',
            num=1, # This number is not currently used
            description=description)

def _add_column_to_spec(spec, column_name, supercolumn, subcolumn,
                        column_type, num, description, path=None):
    """Add a column and associated attributes to the header specification.
    A column includes both a supercolumn (e.g. "SAMPLE") and subcolumn (e.g.
    "DP").
    """
    if supercolumn.upper() not in spec:
        spec[supercolumn.upper()] = OrderedDict()

    spec[supercolumn.upper()][subcolumn] = {
        'path': path if path is not None else [supercolumn, subcolumn],
        'columnName': column_name,
        'name': subcolumn,
        'info': {
            'type': column_type,
            'number': num,
            'description': description
        }
    }


def _find_column(spec, column_name):
    """Returns data for the column in the spec, or None."""
    for columns in spec.itervalues():
        for info in columns.itervalues():
            if info['columnName'] == column_name:
                return info
    return None


def _annotate_query_with_types(query, vcf_spec):
    """Adds a columnType field to each filter and sortBy in query."""
    operations = query.get('sortBy', []) + query.get('filters', [])
    for operation in operations:
        name = operation['columnName']
        info = _find_column(vcf_spec, name)
        if not info:
            if name == 'position':
                operation['columnType'] = 'integer'
            else:
                operation['columnType'] = 'string'
            continue
        column_type = info.get('info', {}).get('type', 'string').lower()
        operation['columnType'] = column_type
    return query


def _whitelist_query_filters(
        query, ok_fields=['reference', 'alternates', 'sample_name']):
    """Returns query with only validation-appropriate filters.

    These are, by default: ['reference', 'alternates', 'sample_name']"""
    query = copy.deepcopy(query)
    query['filters'] = [f for f in query['filters']
                        if f.get('columnName') in ok_fields]
    return query


def _get_vcf_by_id(vcf_id):
    with tables(db.engine, 'vcfs') as (con, vcfs):
        vcf = con.execute(select([vcfs]).where(vcfs.c.id == vcf_id)).fetchone()
    return dict(vcf)