def extract_features(self, idx, pattern_name, pattern_start, pattern_end,
pattern, strand):
task = self.profile_mapping[pattern_name]
ref_profile = self.ref_profiles[pattern_name][task]
narrow_seqlet = Seqlet(idx,
pattern_start,
pattern_end,
name=pattern,
strand=strand)
wide_seqlet = narrow_seqlet.resize(70)
ref_preds = self.dalt_preds[task][idx] # all predictions
ref_preds_seqlet = wide_seqlet.extract(self.dalt_preds[task])
ref_preds_inside = ref_preds_seqlet.sum()
ref_preds_total = ref_preds.sum()
ref_preds_outside = ref_preds_total - ref_preds_inside
try:
ref_preds_match = symmetric_kl(
ref_preds_seqlet,
ref_profile).mean() # compare with the reference
except Exception:
ref_preds_match = np.nan
return {
"pred": {
"inside": ref_preds_inside,
"outside": ref_preds_outside,
"total": ref_preds_total,
"match": ref_preds_match
}
}
def dfi_row2seqlet(row, motifs_inv):
seqlet = Seqlet(row.example_idx,
row.pattern_start,
row.pattern_end,
name=motifs_inv[shorten_te_pattern(row.pattern)],
strand=row.strand)
seqlet.alpha = row.match_weighted_p
return seqlet
def dfi_row2seqlet(row, short_name=False):
return Seqlet(
row.example_idx,
row.pattern_start,
row.pattern_end,
name=shorten_pattern(row.pattern) if short_name else row.pattern,
strand=row.strand)
def fn(dist):
position = dist + 500
sstart, send = motif_coords(side_motif, position)
seqlets = [
Seqlet(None, cstart, cend, "center", ""),
Seqlet(None, sstart, send, "side", "")
]
# TODO - add also importance scores
du = {"p": p[position], "imp": imp[position]}
# TODO - order them correctly
d = OrderedDict([(f"{prefix}/{task}", du[prefix][task])
for task in p[position]
for prefix in ['p', 'imp']])
ylims = []
for k in d:
if k.startswith("p"):
ylims.append((0, ymax))
else:
ylims.append((0, ymax_imp))
plot_tracks(d, seqlets, title=dist, ylim=ylims)
def __getitem__(self, idx):
pair = self.dfab.iloc[idx]
narrow_seqlet_x = Seqlet(seqname=pair.example_idx,
start=pair.pattern_start_x,
end=pair.pattern_end_x,
name="",
strand=pair.strand_x)
narrow_seqlet_y = Seqlet(seqname=pair.example_idx,
start=pair.pattern_start_y,
end=pair.pattern_end_y,
name="",
strand=pair.strand_y)
return {
"x":
self.extract(narrow_seqlet_x,
this_row_idx=pair.row_idx_x,
other_row_idx=pair.row_idx_y,
motif_pair_idx=pair.motif_pair_idx),
"y":
self.extract(narrow_seqlet_y,
this_row_idx=pair.row_idx_y,
other_row_idx=pair.row_idx_x,
motif_pair_idx=pair.motif_pair_idx)
}
def get_change(self, mutated_seqlet_idx, signal_seqlet_idx):
inst = self.dfi.iloc[signal_seqlet_idx]
assert inst.row_idx == signal_seqlet_idx
task = self.profile_mapping[inst.pattern_name]
ref_profile = self.ref_profiles[inst.pattern_name][task]
mutated_inst = self.dfi.iloc[mutated_seqlet_idx]
assert inst.example_idx == mutated_inst.example_idx
narrow_seqlet = Seqlet(inst.example_idx,
inst.pattern_start,
inst.pattern_end,
name=inst.pattern,
strand=inst.strand)
wide_seqlet = narrow_seqlet.resize(70)
# ref
ref_preds = self.preds[task][inst.example_idx] # all predictions
ref_preds_seqlet = wide_seqlet.extract(self.preds[task])
ref_preds_inside = ref_preds_seqlet.sum()
ref_preds_outside = ref_preds.sum() - ref_preds_inside
ref_obs = self.profiles[task][inst.example_idx] # all predictions
ref_obs_seqlet = wide_seqlet.extract(self.profiles[task])
ref_obs_inside = ref_obs_seqlet.sum()
ref_obs_outside = ref_obs.sum() - ref_obs_inside
try:
ref_preds_match = symmetric_kl(
ref_preds_seqlet,
ref_profile).mean() # compare with the reference
except Exception:
ref_preds_match = np.nan
# ref imp
ref_imp_scores = self.ref_imp_scores_contrib[f"{task}/weighted"][
inst.example_idx]
ref_imp_scores_seqlet = narrow_seqlet.extract(
self.ref_imp_scores_contrib[f"{task}/weighted"])
ref_imp_inside = ref_imp_scores_seqlet.sum(
) # sum in the seqlet region
ref_imp_outside = ref_imp_scores.sum(
) - ref_imp_inside # total - seqlet
# ref imp counts
ref_imp_scores_c = self.ref_imp_scores_contrib[f"{task}/count"][
inst.example_idx]
ref_imp_scores_seqlet_c = narrow_seqlet.extract(
self.ref_imp_scores_contrib[f"{task}/count"])
ref_imp_inside_c = ref_imp_scores_seqlet_c.sum(
) # sum in the seqlet region
ref_imp_outside_c = ref_imp_scores_c.sum(
) - ref_imp_inside_c # total - seqlet
# alt
narrow_seqlet.seqname = mutated_seqlet_idx # change the sequence name
wide_seqlet.seqname = mutated_seqlet_idx
alt_preds = self.alt_preds[task][mutated_seqlet_idx]
alt_preds_seqlet = wide_seqlet.extract(self.alt_preds[task])
alt_preds_inside = alt_preds_seqlet.sum() # sum in the seqlet region
alt_preds_outside = alt_preds.sum(
) - alt_preds_inside # total - seqlet
try:
alt_preds_match = symmetric_kl(alt_preds_seqlet,
ref_profile).mean()
except Exception:
alt_preds_match = np.nan
alt_imp_scores = self.alt_imp_scores_contrib[f"{task}/weighted"][
mutated_seqlet_idx]
alt_imp_scores_seqlet = narrow_seqlet.extract(
self.alt_imp_scores_contrib[f"{task}/weighted"])
alt_imp_inside = alt_imp_scores_seqlet.sum(
) # sum in the seqlet region
alt_imp_outside = alt_imp_scores.sum(
) - alt_imp_inside # total - seqlet
alt_imp_scores_c = self.alt_imp_scores_contrib[f"{task}/count"][
mutated_seqlet_idx]
alt_imp_scores_seqlet_c = narrow_seqlet.extract(
self.alt_imp_scores_contrib[f"{task}/count"])
alt_imp_inside_c = alt_imp_scores_seqlet_c.sum(
) # sum in the seqlet region
alt_imp_outside_c = alt_imp_scores_c.sum(
) - alt_imp_inside_c # total - seqlet
return {
"ref": {
"obs": {
"inside": ref_obs_inside,
"outside": ref_obs_outside
},
"pred": {
"inside": ref_preds_inside,
"outside": ref_preds_outside,
"match": ref_preds_match
},
"imp": {
"inside": ref_imp_inside,
"outside": ref_imp_outside,
},
"impcount": {
"inside": ref_imp_inside_c,
"outside": ref_imp_outside_c,
},
},
"alt": {
"pred": {
"inside": alt_preds_inside,
"outside": alt_preds_outside,
"match": alt_preds_match
},
"imp": {
"inside": alt_imp_inside,
"outside": alt_imp_outside,
},
"impcount": {
"inside": alt_imp_inside_c,
"outside": alt_imp_outside_c,
},
},
}