output/preprocess/Colon_and_Rectal_Cancer/code/TCGA.py

# Path Configuration
from tools.preprocess import *

# Processing context
trait = "Colon_and_Rectal_Cancer"

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

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


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

# 1) Select the most appropriate TCGA cohort directory for Colon and Rectal Cancer
subdirs = os.listdir(tcga_root_dir)
preferred_patterns = [
    "TCGA_Colon_and_Rectal_Cancer_(COADREAD)",
    "COADREAD",
    "Colon_and_Rectal_Cancer",
]
selected_dir = None
for pat in preferred_patterns:
    candidates = [d for d in subdirs if pat.lower() in d.lower()]
    if candidates:
        # If multiple options exist, choose the most specific match (first by our preference order)
        selected_dir = sorted(candidates, key=len)[0]
        break

if selected_dir is None:
    # No suitable directory found -> mark as completed and skip
    validate_and_save_cohort_info(
        is_final=False,
        cohort="TCGA_Colon_and_Rectal_Cancer_NotFound",
        info_path=json_path,
        is_gene_available=False,
        is_trait_available=False
    )
else:
    cohort_dir = os.path.join(tcga_root_dir, selected_dir)

    # 2) Identify clinical and genetic file paths
    clinical_file_path, genetic_file_path = tcga_get_relevant_filepaths(cohort_dir)

    # 3) Load both files as DataFrames
    clinical_df = pd.read_csv(clinical_file_path, sep='\t', index_col=0, low_memory=False, compression='infer')
    genetic_df = pd.read_csv(genetic_file_path, sep='\t', index_col=0, low_memory=False, compression='infer')

    # 4) Print the column names of the clinical data
    print(list(clinical_df.columns))

# Step 2: Find Candidate Demographic Features
# Use available clinical_df columns if present; otherwise fall back to the provided list
provided_columns = ['AWG_MLH1_silencing', 'AWG_cancer_type_Oct62011', 'CDE_ID_3226963', 'CIMP', 'MSI_updated_Oct62011', '_INTEGRATION', '_PANCAN_CNA_PANCAN_K8', '_PANCAN_Cluster_Cluster_PANCAN', '_PANCAN_DNAMethyl_COADREAD', '_PANCAN_DNAMethyl_PANCAN', '_PANCAN_RPPA_PANCAN_K8', '_PANCAN_UNC_RNAseq_PANCAN_K16', '_PANCAN_miRNA_PANCAN', '_PANCAN_mutation_PANCAN', '_PATIENT', '_cohort', '_primary_disease', '_primary_site', 'additional_pharmaceutical_therapy', 'additional_radiation_therapy', 'age_at_initial_pathologic_diagnosis', 'anatomic_neoplasm_subdivision', 'bcr_followup_barcode', 'bcr_patient_barcode', 'bcr_sample_barcode', 'braf_gene_analysis_performed', 'braf_gene_analysis_result', 'circumferential_resection_margin', 'colon_polyps_present', 'days_to_birth', 'days_to_collection', 'days_to_death', 'days_to_initial_pathologic_diagnosis', 'days_to_last_followup', 'days_to_new_tumor_event_additional_surgery_procedure', 'days_to_new_tumor_event_after_initial_treatment', 'disease_code', 'followup_case_report_form_submission_reason', 'followup_treatment_success', 'form_completion_date', 'gender', 'height', 'histological_type', 'history_of_colon_polyps', 'history_of_neoadjuvant_treatment', 'hypermutation', 'icd_10', 'icd_o_3_histology', 'icd_o_3_site', 'informed_consent_verified', 'initial_weight', 'intermediate_dimension', 'is_ffpe', 'kras_gene_analysis_performed', 'kras_mutation_codon', 'kras_mutation_found', 'longest_dimension', 'loss_expression_of_mismatch_repair_proteins_by_ihc', 'loss_expression_of_mismatch_repair_proteins_by_ihc_result', 'lost_follow_up', 'lymph_node_examined_count', 'lymphatic_invasion', 'microsatellite_instability', 'new_neoplasm_event_type', 'new_tumor_event_additional_surgery_procedure', 'new_tumor_event_after_initial_treatment', 'non_nodal_tumor_deposits', 'non_silent_mutation', 'non_silent_rate_per_Mb', 'number_of_abnormal_loci', 'number_of_first_degree_relatives_with_cancer_diagnosis', 'number_of_loci_tested', 'number_of_lymphnodes_positive_by_he', 'number_of_lymphnodes_positive_by_ihc', 'oct_embedded', 'pathologic_M', 'pathologic_N', 'pathologic_T', 'pathologic_stage', 'pathology_report_file_name', 'patient_id', 'perineural_invasion_present', 'person_neoplasm_cancer_status', 'postoperative_rx_tx', 'preoperative_pretreatment_cea_level', 'primary_lymph_node_presentation_assessment', 'primary_therapy_outcome_success', 'project_code', 'radiation_therapy', 'residual_disease_post_new_tumor_event_margin_status', 'residual_tumor', 'sample_type', 'sample_type_id', 'shortest_dimension', 'silent_mutation', 'silent_rate_per_Mb', 'site_of_additional_surgery_new_tumor_event_mets', 'synchronous_colon_cancer_present', 'system_version', 'tissue_prospective_collection_indicator', 'tissue_retrospective_collection_indicator', 'tissue_source_site', 'total_mutation', 'tumor_tissue_site', 'venous_invasion', 'vial_number', 'vital_status', 'weight', 'year_of_initial_pathologic_diagnosis', '_GENOMIC_ID_TCGA_COADREAD_exp_HiSeqV2_exon', '_GENOMIC_ID_TCGA_COADREAD_PDMRNAseq', '_GENOMIC_ID_TCGA_COADREAD_exp_HiSeqV2_PANCAN', '_GENOMIC_ID_TCGA_COADREAD_hMethyl450', '_GENOMIC_ID_TCGA_COADREAD_gistic2thd', '_GENOMIC_ID_TCGA_COADREAD_hMethyl27', '_GENOMIC_ID_TCGA_COADREAD_G4502A_07_3', '_GENOMIC_ID_TCGA_COADREAD_PDMarrayCNV', '_GENOMIC_ID_TCGA_COADREAD_exp_HiSeqV2', '_GENOMIC_ID_TCGA_COADREAD_PDMarray', '_GENOMIC_ID_TCGA_COADREAD_gistic2', '_GENOMIC_ID_TCGA_COADREAD_mutation', '_GENOMIC_ID_TCGA_COADREAD_RPPA_RBN', '_GENOMIC_ID_TCGA_COADREAD_PDMRNAseqCNV']

all_columns = list(clinical_df.columns) if 'clinical_df' in globals() else provided_columns

# Identify candidate columns with careful patterns to avoid false positives like "pathologic_stage"
candidate_age_cols = []
candidate_gender_cols = []

for col in all_columns:
    col_l = col.lower()
    # Age candidates
    if (
        col_l == 'age' or
        col_l.startswith('age_') or
        'age_at' in col_l or
        col_l.endswith('_age') or
        col_l in {'days_to_birth', 'years_to_birth', 'year_of_birth', 'dob'}
    ):
        candidate_age_cols.append(col)
    # Gender candidates (avoid partial matches like "seq")
    if (
        col_l in {'gender', 'sex'} or
        col_l.endswith('_gender') or
        col_l.endswith('_sex') or
        col_l.startswith('gender_') or
        col_l.startswith('sex_')
    ):
        candidate_gender_cols.append(col)

# 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 if clinical_df is available
if 'clinical_df' in globals():
    age_cols_present = [c for c in candidate_age_cols if c in clinical_df.columns]
    gender_cols_present = [c for c in candidate_gender_cols if c in clinical_df.columns]

    if age_cols_present:
        print(preview_df(clinical_df[age_cols_present], n=5))
    if gender_cols_present:
        print(preview_df(clinical_df[gender_cols_present], n=5))

# Step 3: Select Demographic Features
import pandas as pd
import numpy as np

# Defaults
age_col = None
gender_col = None

# Heuristics thresholds
min_non_missing_ratio = 0.6

# Helper to check if a variable exists
def var_exists(name):
    return name in globals() or name in locals()

# Select age column
if var_exists('candidate_age_cols'):
    if 'clinical_df' in globals() or 'clinical_df' in locals():
        df = clinical_df
        best_score = -np.inf
        best_col = None
        for col in candidate_age_cols:
            if col in df.columns:
                s = pd.to_numeric(df[col], errors='coerce')
                non_missing_ratio = s.notna().mean()

                # Plausible human age range in years
                plausible_ratio = ((s >= 0) & (s <= 120)).mean(skipna=True)

                # Small bonus if column name suggests age in years
                name_bonus = 0.1 if 'age' in col.lower() and 'birth' not in col.lower() else 0.0

                score = plausible_ratio * 1.0 + non_missing_ratio * 0.2 + name_bonus

                if non_missing_ratio >= min_non_missing_ratio and score > best_score:
                    best_score = score
                    best_col = col

        age_col = best_col
    else:
        # Fallback to commonly correct choice if DataFrame not available
        age_col = 'age_at_initial_pathologic_diagnosis' if 'age_at_initial_pathologic_diagnosis' in candidate_age_cols else None

# Select gender column
if var_exists('candidate_gender_cols'):
    if 'clinical_df' in globals() or 'clinical_df' in locals():
        df = clinical_df
        best_score = -np.inf
        best_col = None

        allowed = {'male', 'female', 'm', 'f'}
        for col in candidate_gender_cols:
            if col in df.columns:
                s = df[col].astype(str).str.strip().str.lower()
                non_missing_ratio = df[col].notna().mean()
                in_allowed = s.isin(allowed)
                allowed_ratio = in_allowed.mean()

                # Score prioritizes valid gender values and completeness
                score = allowed_ratio * 1.0 + non_missing_ratio * 0.2

                if non_missing_ratio >= min_non_missing_ratio and score > best_score:
                    best_score = score
                    best_col = col

        gender_col = best_col
    else:
        gender_col = 'gender' if 'gender' in candidate_gender_cols else None

# Print selected columns and a brief preview if available
print("Selected age_col:", age_col)
if age_col is not None and ('clinical_df' in globals() or 'clinical_df' in locals()):
    print("age_col preview (first 5):", clinical_df[age_col].head(5).tolist())

print("Selected gender_col:", gender_col)
if gender_col is not None and ('clinical_df' in globals() or 'clinical_df' in locals()):
    print("gender_col preview (first 5):", clinical_df[gender_col].head(5).tolist())

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

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

# 2) Prepare genetic data with genes as index, samples as columns
def _tcga_prop_tcga_prefix(labels):
    if len(labels) == 0:
        return 0.0
    return np.mean([isinstance(x, str) and x.startswith('TCGA') for x in labels])

# Detect orientation: are TCGA sample IDs in index or columns?
p_idx = _tcga_prop_tcga_prefix(genetic_df.index.tolist())
p_col = _tcga_prop_tcga_prefix(genetic_df.columns.tolist())

if p_idx >= 0.5 and p_idx > p_col:
    # Index are samples; transpose to get genes as index
    gene_df_raw = genetic_df.T
else:
    gene_df_raw = genetic_df

# Ensure numeric and drop all-nan rows/cols safely
gene_df_raw = gene_df_raw.apply(pd.to_numeric, errors='coerce')
gene_df_raw = gene_df_raw.dropna(axis=0, how='all').dropna(axis=1, how='all')

# Normalize gene symbols using NCBI synonyms and aggregate duplicates
normalized_gene_df = normalize_gene_symbols_in_index(gene_df_raw)

# Save normalized gene expression data
os.makedirs(os.path.dirname(out_gene_data_file), exist_ok=True)
normalized_gene_df.to_csv(out_gene_data_file)

# 3) Link clinical and genetic data on sample IDs
# Harmonize sample identifiers to first 15 chars (e.g., TCGA-XX-XXXX-01)
def _to_sample15(s):
    return str(s)[:15] if isinstance(s, str) else s

E = normalized_gene_df.T.copy()  # samples x genes
E.index = E.index.map(_to_sample15)
E = E[~E.index.duplicated(keep='first')]

clinical_harmonized = selected_clinical_df.copy()
clinical_harmonized.index = clinical_harmonized.index.map(_to_sample15)
clinical_harmonized = clinical_harmonized[~clinical_harmonized.index.duplicated(keep='first')]

linked_data = clinical_harmonized.join(E, how='inner')

# 4) Handle missing values systematically
processed_df = handle_missing_values(linked_data, trait_col=trait)

# 5) Determine bias in trait and demographics; remove biased demographics
is_biased, debiased_df = judge_and_remove_biased_features(processed_df, trait=trait)
is_biased = bool(is_biased)  # ensure Python-native bool

# 6) Final quality validation and save cohort metadata
cohort_name = selected_dir if 'selected_dir' in globals() else "TCGA_Colon_and_Rectal_Cancer_(COADREAD)"
is_gene_available = bool((normalized_gene_df.shape[0] > 0) and (normalized_gene_df.shape[1] > 0))
is_trait_available = bool((trait in debiased_df.columns) and bool(debiased_df[trait].notna().any()))

note = (
    f"INFO: Linked clinical and gene expression data for {cohort_name}. "
    f"Normalized genes: {normalized_gene_df.shape[0]}; samples in gene data: {normalized_gene_df.shape[1]}. "
    f"Linked samples after harmonization: {linked_data.shape[0]}; final samples after QC: {debiased_df.shape[0]}; "
    f"final features: {debiased_df.shape[1]}."
)

is_usable = validate_and_save_cohort_info(
    is_final=True,
    cohort=str(cohort_name),
    info_path=json_path,
    is_gene_available=is_gene_available,
    is_trait_available=is_trait_available,
    is_biased=is_biased,
    df=debiased_df,
    note=note
)

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