output/preprocess/Breast_Cancer/code/TCGA.py

# Path Configuration
from tools.preprocess import *

# Processing context
trait = "Breast_Cancer"

# Input paths
tcga_root_dir = "../DATA/TCGA"

# Output paths
out_data_file = "./output/z2/preprocess/Breast_Cancer/TCGA.csv"
out_gene_data_file = "./output/z2/preprocess/Breast_Cancer/gene_data/TCGA.csv"
out_clinical_data_file = "./output/z2/preprocess/Breast_Cancer/clinical_data/TCGA.csv"
json_path = "./output/z2/preprocess/Breast_Cancer/cohort_info.json"


# Step 1: Initial Data Loading
import os
import pandas as pd

# Step 1: Select the most relevant TCGA cohort directory for Breast Cancer
def select_tcga_cohort(root_dir: str, trait_keywords=None):
    if trait_keywords is None:
        trait_keywords = ['breast', 'brca']

    subdirs = [d for d in os.listdir(root_dir) if os.path.isdir(os.path.join(root_dir, d))]
    best_dir, best_score = None, -1
    for d in subdirs:
        name = d.lower()
        score = 0
        if 'breast' in name:
            score += 5
        if 'brca' in name:
            score += 3
        if 'tcga_breast_cancer' in name:
            score += 10
        if score > best_score:
            best_score = score
            best_dir = d
    return best_dir, best_score

selected_dir, score = select_tcga_cohort(tcga_root_dir)
if (selected_dir is None) or (score <= 0):
    # No appropriate cohort found; record and stop early for this trait
    validate_and_save_cohort_info(
        is_final=False,
        cohort="TCGA",
        info_path=json_path,
        is_gene_available=False,
        is_trait_available=False
    )
else:
    cohort_dir = os.path.join(tcga_root_dir, selected_dir)
    print(f"Selected cohort directory: {selected_dir}")

    # Step 2: Identify clinical and genetic data file paths
    clinical_file_path, genetic_file_path = tcga_get_relevant_filepaths(cohort_dir)
    print(f"Clinical file: {os.path.basename(clinical_file_path)}")
    print(f"Genetic file: {os.path.basename(genetic_file_path)}")

    # Step 3: Load both files as DataFrames
    def read_tcga_file(path: str) -> pd.DataFrame:
        compression = 'gzip' if path.endswith('.gz') else None
        return pd.read_csv(path, sep='\t', index_col=0, low_memory=False, compression=compression)

    clinical_df = read_tcga_file(clinical_file_path)
    genetic_df = read_tcga_file(genetic_file_path)

    # Step 4: Print clinical column names
    print(list(clinical_df.columns))

# Step 2: Find Candidate Demographic Features
import os
import re
import pandas as pd

# Determine paths
cohort_dir_name = "TCGA_Breast_Cancer_(BRCA)"
clinical_file_name = "TCGA.BRCA.sampleMap_BRCA_clinicalMatrix"
cohort_dir = os.path.join(tcga_root_dir, cohort_dir_name)
clinical_path = os.path.join(cohort_dir, clinical_file_name)

# Load clinical data
clinical_df = pd.read_csv(clinical_path, sep="\t", index_col=0, dtype=str)

# Identify candidate columns
def is_age_col(col: str) -> bool:
    cl = col.lower()
    if 'days_to_birth' in cl:
        return True
    if 'age_at' in cl:
        return True
    if re.search(r'(^|_)age(_|$)', cl):
        return True
    return False

def is_gender_col(col: str) -> bool:
    cl = col.lower()
    if re.search(r'(^|_)gender(_|$)', cl):
        return True
    if re.search(r'(^|_)sex(_|$)', cl):
        return True
    return False

cols = list(clinical_df.columns)
candidate_age_cols = [c for c in cols if is_age_col(c)]
candidate_gender_cols = [c for c in cols if is_gender_col(c)]

# Print required lists in strict format
print(f"candidate_age_cols = {candidate_age_cols}")
print(f"candidate_gender_cols = {candidate_gender_cols}")

# Preview extracted data dictionaries
if candidate_age_cols:
    age_df = clinical_df[candidate_age_cols]
    print(preview_df(age_df, n=5))
if candidate_gender_cols:
    gender_df = clinical_df[candidate_gender_cols]
    print(preview_df(gender_df, n=5))

# Step 3: Select Demographic Features
# Select demographic feature columns based on preview dictionaries if available; fallback to sensible defaults.

age_col = None
gender_col = None

def _safe_get(var_name, default=None):
    # Access a variable by name if it exists in globals()
    return globals().get(var_name, default)

# Attempt to retrieve the preview dictionaries from previous steps
age_preview_dict = _safe_get('age_preview_dict', None)
gender_preview_dict = _safe_get('gender_preview_dict', None)

def is_valid_age_value(v):
    a = tcga_convert_age(v)
    return a is not None and 0 <= a <= 120

def is_valid_gender_value(v):
    g = tcga_convert_gender(v)
    return g is not None and g in (0, 1)

# Determine age_col
if isinstance(age_preview_dict, dict) and isinstance(globals().get('candidate_age_cols', None), list):
    best_col = None
    best_score = -1
    for c in candidate_age_cols:
        vals = age_preview_dict.get(c, [])
        valid_count = sum(1 for v in vals if is_valid_age_value(v))
        # Heuristic tie-breakers: prefer explicit age columns over 'days_to_birth', and avoid 'nature2012' if tied
        score = valid_count
        if 'days_to_birth' in c.lower():
            score -= 0.5
        if 'nature2012' in c.lower():
            score -= 0.1
        if score > best_score:
            best_score = score
            best_col = c
    age_col = best_col if best_score > 0 else None
else:
    # Fallback selection if preview dict not available: prioritize typical age column names
    if 'age_at_initial_pathologic_diagnosis' in globals().get('candidate_age_cols', []):
        age_col = 'age_at_initial_pathologic_diagnosis'
    elif 'Age_at_Initial_Pathologic_Diagnosis_nature2012' in globals().get('candidate_age_cols', []):
        age_col = 'Age_at_Initial_Pathologic_Diagnosis_nature2012'
    elif 'days_to_birth' in globals().get('candidate_age_cols', []):
        age_col = 'days_to_birth'
    else:
        age_col = None

# Determine gender_col
if isinstance(gender_preview_dict, dict) and isinstance(globals().get('candidate_gender_cols', None), list):
    best_col = None
    best_score = -1
    for c in candidate_gender_cols:
        vals = gender_preview_dict.get(c, [])
        valid_count = sum(1 for v in vals if is_valid_gender_value(v))
        # Prefer columns without 'nature2012' if tied
        score = valid_count - (0.1 if 'nature2012' in c.lower() else 0)
        if score > best_score:
            best_score = score
            best_col = c
    gender_col = best_col if best_score > 0 else None
else:
    # Fallback selection if preview dict not available: prefer standard 'gender'
    if 'gender' in globals().get('candidate_gender_cols', []):
        gender_col = 'gender'
    elif 'Gender_nature2012' in globals().get('candidate_gender_cols', []):
        gender_col = 'Gender_nature2012'
    else:
        gender_col = None

# Explicitly print out the information for the chosen columns
print(f"Selected age_col: {age_col}")
if age_col is not None and isinstance(age_preview_dict, dict) and age_col in age_preview_dict:
    print(f"Preview values for age_col ({age_col}): {age_preview_dict[age_col]}")

print(f"Selected gender_col: {gender_col}")
if gender_col is not None and isinstance(gender_preview_dict, dict) and gender_col in gender_preview_dict:
    print(f"Preview values for gender_col ({gender_col}): {gender_preview_dict[gender_col]}")

# Step 4: Feature Engineering and Validation
import os
import pandas as pd

# 1) Extract and standardize clinical features (trait, Age, Gender)
selected_clinical_df = tcga_select_clinical_features(
    clinical_df=clinical_df,
    trait=trait,
    age_col=age_col,
    gender_col=gender_col
)

# Helper to detect TCGA sample barcode
def _is_tcga_barcode(x: str) -> bool:
    return isinstance(x, str) and x.startswith("TCGA-")

def _fraction_tcga(seq) -> float:
    if len(seq) == 0:
        return 0.0
    cnt = sum(1 for v in seq if _is_tcga_barcode(str(v)))
    return cnt / len(seq)

# 2) Normalize gene symbols in gene expression data
is_gene_available = False
normalized_gene_df = pd.DataFrame()
gene_note = ""

try:
    if isinstance(genetic_df, pd.DataFrame) and len(genetic_df) > 0:
        idx_frac = _fraction_tcga(list(genetic_df.index[:min(len(genetic_df.index), 1000)]))
        col_frac = _fraction_tcga(list(genetic_df.columns[:min(len(genetic_df.columns), 1000)]))

        # Orient to genes as index, samples as columns
        if idx_frac > 0.5 and col_frac <= 0.5:
            oriented = genetic_df.T
            gene_note += "INFO: Genetic data transposed to have genes as index and samples as columns. "
        elif col_frac > 0.5 and idx_frac <= 0.5:
            oriented = genetic_df.copy()
            gene_note += "INFO: Genetic data already oriented with genes as index and samples as columns. "
        else:
            oriented = genetic_df.copy()
            gene_note += "WARNING: Genetic data orientation ambiguous. Proceeded without transposition. "

        # Coerce to numeric
        oriented = oriented.apply(pd.to_numeric, errors='coerce')
        # Keep columns that look like TCGA barcodes to ensure samples are columns
        sample_cols = [c for c in oriented.columns if _is_tcga_barcode(c)]
        if len(sample_cols) == 0:
            gene_note += "ERROR: No sample barcode-like columns detected in genetic data. "
            is_gene_available = False
        else:
            oriented = oriented.loc[:, sample_cols]

            # Normalize gene symbols in index
            normalized_gene_df = normalize_gene_symbols_in_index(oriented)

            # Heuristic checks for availability
            genes_count = normalized_gene_df.shape[0]
            samples_count = normalized_gene_df.shape[1]
            has_values = normalized_gene_df.notna().any().any()
            is_gene_available = (genes_count >= 500) and (samples_count >= 10) and has_values

            # Save normalized gene data if available
            if bool(is_gene_available):
                os.makedirs(os.path.dirname(out_gene_data_file), exist_ok=True)
                normalized_gene_df.to_csv(out_gene_data_file)
            else:
                gene_note += f"ERROR: Insufficient gene expression data after normalization (genes={genes_count}, samples={samples_count}). "
    else:
        gene_note += "ERROR: Genetic dataframe missing or empty. "
except Exception as e:
    gene_note += f"ERROR: Exception during gene processing: {e}. "
    is_gene_available = False
    normalized_gene_df = pd.DataFrame()

# 3) Link clinical and genetic data
if bool(is_gene_available):
    common_samples = selected_clinical_df.index.intersection(normalized_gene_df.columns)
    linked_data = pd.concat(
        [selected_clinical_df.loc[common_samples], normalized_gene_df.loc[:, common_samples].T],
        axis=1
    )
else:
    linked_data = selected_clinical_df.copy()

# 4) Handle missing values
covariate_cols = [trait, "Age", "Gender"]
gene_cols_in_linked = [c for c in linked_data.columns if c not in covariate_cols]
if len(gene_cols_in_linked) > 0:
    linked_data = handle_missing_values(linked_data, trait)
else:
    linked_data = linked_data.dropna(subset=[trait])
    if "Age" in linked_data.columns:
        linked_data["Age"] = pd.to_numeric(linked_data["Age"], errors="coerce")
        linked_data["Age"] = linked_data["Age"].fillna(linked_data["Age"].mean())
    if "Gender" in linked_data.columns:
        mode_result = linked_data["Gender"].mode()
        if len(mode_result) > 0:
            linked_data["Gender"] = linked_data["Gender"].fillna(mode_result[0])
        else:
            linked_data = linked_data.drop(columns=["Gender"])

# 5) Determine biased features and remove biased demographics
trait_biased, linked_data = judge_and_remove_biased_features(linked_data, trait)

# 6) Final validation and save metadata
# Recompute trait availability on the linked dataset to reflect post-processing status
is_trait_available = bool(linked_data[trait].notna().any())
note = gene_note.strip() if gene_note else ""
note = note if note.startswith(("INFO:", "WARNING:", "ERROR:", "DEBUG:")) else (("INFO: " + note) if note else "")

is_usable = validate_and_save_cohort_info(
    is_final=True,
    cohort="TCGA",
    info_path=json_path,
    is_gene_available=bool(is_gene_available),
    is_trait_available=bool(is_trait_available),
    is_biased=bool(trait_biased),
    df=linked_data,
    note=note
)

# 7) Save linked data if usable
if bool(is_usable):
    os.makedirs(os.path.dirname(out_data_file), exist_ok=True)
    linked_data.to_csv(out_data_file)