def test_scale_dataframe():
df = pd.DataFrame({"A": [1, 2, 3, 4], "B": [10, 20, 30, 40]})
#If scalled, A and B should be same
df_sum = list(core.scale_dataframe(df).sum())
assert df_sum[0] == df_sum[1]
def create_inter_samples_model(self):
"""Create inter sample model
This model compute the mean and the standard deviation from training sampling.
This will be used later to compute a inter z-score in a new sample
"""
self.norm_raw = core.scale_dataframe(self.raw)
self.inter_model = pd.DataFrame({
"mean": self.norm_raw.mean(axis=1),
"median": self.norm_raw.median(axis=1),
"std": self.norm_raw.std(axis=1),
"min": self.norm_raw.min(axis=1),
"max": self.norm_raw.max(axis=1),
})
def test_sample(self, bamfile: str) -> pd.DataFrame:
"""Test a new sample against the current model
model = Model("model.h5")
data = model.test_sample("sample.bam")
The dataframe contains for each position:
- depth : the raw depth
- depth_norm: the normalized raw depth
- inter_z: the inter-model z-score
- depth_mate: the raw depth of the mate
- depth_mate_predicted: the raw depth predicted by the intra-model
- intra_z: the intra-model z-score
Args:
bamfile (str): A sample bam file
Returns:
pd.DataFrame
"""
del_coverage = core.get_coverages_from_bed([bamfile], self.bedfile)
dd = del_coverage.copy()
dd.columns = ["depth"]
# Compute inter model
dd["depth_norm"] = core.scale_dataframe(dd)["depth"]
dd["inter_z"] = (dd["depth_norm"] -
self.inter_model["mean"]) / self.inter_model["std"]
# Compute intra model
subset = dd.loc[self.intra_model.index]
subset["depth_mate"] = subset.iloc[
self.intra_model["idx"], :].iloc[:, 0].values
subset["depth_mate_predicted"] = (
self.intra_model["coef"] *
subset["depth"]) + self.intra_model["intercept"]
subset["corr"] = self.intra_model["corr"]
subset["error"] = subset["depth_mate_predicted"] - subset["depth_mate"]
subset["intra_z"] = subset["error"] / self.intra_model["std"]
subset.drop(["depth", "depth_norm", "inter_z"], axis=1)
test_data = dd.join(
subset.drop(["depth", "depth_norm", "inter_z"], axis=1))
return test_data
def test_sample(self, bamfile: str, show_progress=True) -> pd.DataFrame:
"""Test a new sample against the current model
model = Model("model.h5")
data = model.test_sample("sample.bam")
The dataframe contains for each position:
- depth : the raw depth
- depth_norm: the normalized raw depth
- inter_z: the inter-model z-score
- depth_mate: the raw depth of the mate
- depth_mate_predicted: the raw depth predicted by the intra-model
- intra_z: the intra-model z-score
Args:
bamfile (str): A sample bam file
Returns:
pd.DataFrame
"""
del_coverage = core.get_coverages_from_bed(
bamfile,
self.bedfile,
sample_rate=self.sample_rate,
show_progress=show_progress,
)
dd = del_coverage.copy()
dd.columns = ["depth"]
# Compute inter model
dd["depth_norm"] = core.scale_dataframe(dd)["depth"]
dd["inter_z"] = (dd["depth_norm"] -
self.super_model["mean"]) / self.super_model["std"]
# # Compute intra model
depth_mate = dd.iloc[self.super_model["idx"], :]["depth"].to_list()
dd["depth_mate"] = depth_mate
dd["depth_mate_predicted"] = (self.super_model["coef"] * dd["depth"]
) + self.super_model["intercept"]
dd["error_intra"] = dd["depth_mate_predicted"] - dd["depth_mate"]
dd["intra_z"] = dd["error_intra"] / self.super_model["std2"]
return dd