Hconsetacc

Function

hconsetacc(
    self,
    tool_hap1_cna_files: List[str],
    tool_hap2_cna_files: List[str],
    tool_names: List[str],
    changes_file: str,
    profile_bin_size=100000,
    outfile: str = "evolution_onset_acc.csv",
) -> pd.DataFrame

This function evaluates evolution onset change classification accuracy for multiple tools.

Given a ground-truth change table (segments annotated with change Type and GT Change) and each tool’s haplotype-specific CNA profiles, it:

1. Re-bins GT change segments and tool CNA profiles to a uniform resolution (profile_bin_size).
2. Joins GT segments with tool predictions for hap1 and hap2.
3. Derives a predicted change label (Change_predict) via your internal onset-join logic.
4. Computes accuracy (ACC) of Change_predict vs GT Change, stratified by change Type.
5. Saves both intermediate combined tables (per tool) and the final summary table.

---

Parameters

Name	Type	Description
tool_hap1_cna_files	List[str]	List of tool CNA profile CSVs for haplotype 1. Must align with tool_names.
tool_hap2_cna_files	List[str]	List of tool CNA profile CSVs for haplotype 2. Must align with tool_names.
tool_names	List[str]	Tool names used for output file naming and result rows.
changes_file	str	Path to the ground-truth change table (segments + labels).
profile_bin_size	int	Bin size used to split/standardize both GT segments and predictions. Default: 100000 (100kb).
outfile	str	Output filename for the summary table. Default: "evolution_onset_acc.csv".

---

Input File Format

changes_file (ground-truth change table)

Expected to contain at least:

• Segment: genomic segment identifier (used as join key after re-binning)
• Type: change category label (used for stratified reporting)
• Change: ground-truth change label used for accuracy evaluation

Optional:

• Haplotype: if present, will be normalized via self._normalize_hap_label(...)

Implementation details:

change_truth = read_and_drop_empty(changes_file)
change_truth = split_all_regions(change_truth.set_index("Segment"), profile_bin_size)
change_truth = change_truth.reset_index().rename(columns={"index": "Segment"})

Tool CNA profiles

Each file is expected to have:

• region column
• remaining columns as cell IDs
• values: haplotype-specific CNA states (format depends on your pipeline)

They are loaded and re-binned as:

p_h1 = split_all_regions(p_h1.set_index("region"), profile_bin_size).reset_index()
p_h2 = split_all_regions(p_h2.set_index("region"), profile_bin_size).reset_index()

---

Evaluation Logic

For each tool:

1. Join GT changes with haplotype predictions:

h1 = self._onset_join(change_truth, p_h1, "hap1")
h2 = self._onset_join(change_truth, p_h2, "hap2")
comb = pd.concat([h1, h2], ignore_index=True)

The joined table is expected to include (at minimum):

• Type
• Change (GT)

• Change_predict (predicted change label)

• Save the per-tool combined table for debugging/inspection:

{self.output_dir}/{ToolName}_comb_combined.csv

1. Compute accuracy per Type:

2. Compare Change_predict vs Change only where both are non-missing:

mask = gt.notna() & pd_.notna()
acc = (pd_[mask] == gt[mask]).mean()

1. Append one result row per (Tool × Type):

2. Tool

3. Type
4. ACC

---

Output

Intermediate per-tool combined tables

For each tool, the function writes:

os.path.join(self.output_dir, f"{name}_comb_combined.csv")

These files contain the merged GT + predictions across both haplotypes.

Final summary table (outfile)

Saved to:

os.path.join(self.output_dir, outfile)

Default:

{self.output_dir}/evolution_onset_acc.csv

---

Output Table Schema

The final CSV / dataframe contains one row per (Tool × Type):

Column	Meaning
Tool	Tool name
Type	Change category label from the GT table
ACC	Accuracy of predicted change labels (Change_predict) vs GT (Change)

Notes:

• If a given Type has no valid comparable entries (all missing), ACC is set to NaN.

---

Return Value

Returns a pd.DataFrame with columns:

• Tool
• Type
• ACC

---

Example

from hcbench.gtbench.gtbench import GTBench

bench = GTBench(output_dir="out/gt_output")

df = bench.hconsetacc(
    tool_hap1_cna_files=[
        "/path/to/chisel/hap1.csv",
        "/path/to/signals/hap1.csv",
    ],
    tool_hap2_cna_files=[
        "/path/to/chisel/hap2.csv",
        "/path/to/signals/hap2.csv",
    ],
    tool_names=["CHISEL", "SIGNALS"],
    changes_file="/path/to/gt/changes_onset.csv",
    profile_bin_size=100000,
    outfile="evolution_onset_acc.csv",
)

print(df)