def select_informative_positions(data, incid_thresh):
minor_allele_incid = (data > 0).mean("library_id").min("allele")
informative_positions = idxwhere(
minor_allele_incid.to_series() > incid_thresh
)
return informative_positions
f"Found {npos_available} informative positions with minor "
f"allele incidence of >{args.incid_thresh}"
)
npos = min(args.npos, npos_available)
info(f"Randomly sampling {npos} positions.")
position_ss = np.random.choice(
informative_positions,
size=npos,
replace=False,
)
info("Filtering libraries.")
suff_cvrg_samples = idxwhere(
(
(
data.sel(position=informative_positions).sum(["allele"]) > 0
).mean("position")
> args.cvrg_thresh
).to_series()
)
nlibs = len(suff_cvrg_samples)
info(
f"Found {nlibs} libraries with >{args.cvrg_thresh:0.1%} "
f"of informative positions covered."
)
info("Constructing input data.")
data_fit = data.sel(library_id=suff_cvrg_samples, position=position_ss)
m_ss = data_fit.sum("allele")
n, g_ss = m_ss.shape
y_obs_ss = data_fit.sel(allele="alt")
cvrg = trsnf_data.sum("allele")
trsnf_data = (trsnf_data + 1) / (cvrg + 2)
trsnf_data = trsnf_data.sel(allele="alt") * 2 - 1
info(f"Clustering {nlibs} metagenotypes at a {dist_thresh} "
f"maximum cosine distance threshold across {npos} positions.")
clust = AgglomerativeClustering(
n_clusters=None,
affinity="cosine",
linkage="complete",
distance_threshold=dist_thresh,
).fit(trsnf_data)
clust = pd.Series(clust.labels_, index=trsnf_data.library_id)
clust_size = clust.value_counts()
large_clusts = idxwhere(clust_size >= clust_size_thresh)
nclusts = len(large_clusts)
total_clust_libs = clust_size.loc[large_clusts].sum()
info(f"Found {nclusts} ('large') clusters with at least "
f"{clust_size_thresh} members, encompassing a total of "
f"{total_clust_libs} libraries.")
info(f"Constructing cluster-by-position matrix indicating "
f"which large clusters have counts at each position in "
f">{clust_pos_frac_thresh:.1%} of samples.")
cvrg_subset = cvrg.to_pandas().loc[clust.index]
clust_pos_frac = {}
for c in large_clusts:
clust_pos_frac[c] = (cvrg_subset.loc[clust == c] > 0).mean()
clust_pos_frac = pd.DataFrame(clust_pos_frac, index=inf_positions)
clust_pos_incid = (clust_pos_frac >= clust_pos_frac_thresh).astype(int)
def xr_idxwhere(x):
return idxwhere(x.to_series())
def filter_samples(pileup, min_median_coverage):
median_coverage = (pileup.groupby(level='sample_id', axis='columns')
.sum()
.median())
return pileup.loc[:, idxwhere(median_coverage >= min_median_coverage)]