class BedReader(object):
def __init__(self, path, shape, dtype=np.int8, count_A1=True):
# n variants (sid = SNP id), n samples (iid = Individual id)
n_sid, n_iid = shape
# Initialize Bed with empty arrays for axis data, otherwise it will
# load the bim/map/fam files entirely into memory (it does not do out-of-core for those)
self.bed = Bed(
str(path),
count_A1=count_A1,
# Array (n_sample, 2) w/ FID and IID
iid=np.empty((n_iid, 2), dtype="str"),
# SNP id array (n_variants)
sid=np.empty((n_sid, ), dtype="str"),
# Contig and positions array (n_variants, 3)
pos=np.empty((n_sid, 3), dtype="int"),
)
self.shape = (n_sid, n_iid, 2)
self.dtype = dtype
self.ndim = 3
def __getitem__(self, idx):
if not isinstance(idx, tuple):
raise IndexError( # pragma: no cover
f"Indexer must be tuple (received {type(idx)})")
if len(idx) != self.ndim:
raise IndexError( # pragma: no cover
f"Indexer must be two-item tuple (received {len(idx)} slices)")
# Slice using reversal of first two slices --
# pysnptools uses sample x variant orientation
arr = self.bed[idx[1::-1]].read(dtype=np.float32, view_ok=False).val.T
# Convert missing calls as nan to -1
arr = np.nan_to_num(arr, nan=-1.0)
arr = arr.astype(self.dtype)
# Add a ploidy dimension, so allele counts of 0, 1, 2 correspond to 00, 10, 11
arr = np.stack(
[
np.where(arr < 0, -1, np.where(arr == 0, 0, 1)),
np.where(arr < 0, -1, np.where(arr == 2, 1, 0)),
],
axis=-1,
)
# Apply final slice to 3D result
return arr[:, :, idx[-1]]
def close(self):
# This is not actually crucial since a Bed instance with no
# in-memory bim/map/fam data is essentially just a file pointer
# but this will still be problematic if the an array is created
# from the same PLINK dataset many times
self.bed._close_bed() # pragma: no cover
class BedReader(object):
def __init__(self, path, shape, dtype=np.int8, count_A1=True):
from pysnptools.snpreader import Bed
# n variants (sid = SNP id), n samples (iid = Individual id)
n_sid, n_iid = shape
# Initialize Bed with empty arrays for axis data, otherwise it will
# load the bim/map/fam files entirely into memory (it does not do out-of-core for those)
self.bed = Bed(
str(path),
count_A1=count_A1,
# Array (n_sample, 2) w/ FID and IID
iid=np.empty((n_iid, 2), dtype='str'),
# SNP id array (n_variants)
sid=np.empty((n_sid, ), dtype='str'),
# Contig and positions array (n_variants, 3)
pos=np.empty((n_sid, 3), dtype='int'))
self.shape = (n_sid, n_iid)
self.dtype = dtype
self.ndim = 2
@staticmethod
def _is_empty_slice(s):
return s.start == s.stop
def __getitem__(self, idx):
if not isinstance(idx, tuple):
raise IndexError(f'Indexer must be tuple (received {type(idx)})')
if len(idx) != self.ndim:
raise IndexError(
f'Indexer must be two-item tuple (received {len(idx)} slices)')
# This is called by dask with empty slices before trying to read any chunks, so it may need
# to be handled separately if pysnptools is slow here
# if all(map(BedReader._is_empty_slice, idx)):
# return np.empty((0, 0), dtype=self.dtype)
arr = self.bed[idx[::-1]].read(dtype=np.float32, view_ok=False).val.T
arr = np.ma.masked_invalid(arr)
arr = arr.astype(self.dtype)
return arr
def close(self):
# This is not actually crucial since a Bed instance with no
# in-memory bim/map/fam data is essentially just a file pointer
# but this will still be problematic if the an array is created
# from the same PLINK dataset many times
self.bed._close_bed()
