p3/preprocess/Thymoma/code/GSE29695.py

# Path Configuration
from tools.preprocess import *

# Processing context
trait = "Thymoma"
cohort = "GSE29695"

# Input paths
in_trait_dir = "../DATA/GEO/Thymoma"
in_cohort_dir = "../DATA/GEO/Thymoma/GSE29695"

# Output paths
out_data_file = "./output/preprocess/3/Thymoma/GSE29695.csv"
out_gene_data_file = "./output/preprocess/3/Thymoma/gene_data/GSE29695.csv"
out_clinical_data_file = "./output/preprocess/3/Thymoma/clinical_data/GSE29695.csv"
json_path = "./output/preprocess/3/Thymoma/cohort_info.json"

# Get file paths
soft_file_path, matrix_file_path = geo_get_relevant_filepaths(in_cohort_dir)

# Get background info and clinical data 
background_info, clinical_data = get_background_and_clinical_data(matrix_file_path)

# Print shape and first few rows to verify data
print("Background Information:")
print(background_info)
print("\nClinical Data Shape:", clinical_data.shape)
print("\nFirst few rows of Clinical Data:")
print(clinical_data.head())

print("\nSample Characteristics:")
# Get dictionary of unique values per row
unique_values_dict = get_unique_values_by_row(clinical_data)
for row, values in unique_values_dict.items():
    print(f"\n{row}:")
    print(values)
# 1. Gene Expression Data Availability
# From the background information, this is a gene expression dataset using Illumina BeadStudio platform
is_gene_available = True

# 2.1 Data Availability
# Trait can be determined from type/category field (row 1 or 2)
trait_row = 1  # Using the type field which has more granular classification

# Age and gender not available
age_row = None 
gender_row = None

# 2.2 Data Type Conversion Functions
def convert_trait(value: str) -> int:
    """Convert thymoma subtype to binary (0=less aggressive, 1=more aggressive)"""
    if not value or ':' not in value:
        return None
    
    type_val = value.split(':')[1].strip()
    
    # B3 is most aggressive subtype
    if 'B3' in type_val:
        return 1
    # B2, B1/B2 are intermediate aggression
    elif 'B2' in type_val or 'B1/B2' in type_val:
        return 1  
    # A, AB, B1 are less aggressive
    elif type_val in ['A', 'AB', 'B1', 'Mixed AB', 'A/B']:
        return 0
    # Cell lines should be excluded
    elif type_val == 'CL':
        return None
    return None

def convert_age(value: str) -> float:
    return None  # Age not available

def convert_gender(value: str) -> int:
    return None  # Gender not available

# 3. Save initial metadata
is_trait_available = trait_row is not None
validate_and_save_cohort_info(is_final=False, 
                            cohort=cohort,
                            info_path=json_path,
                            is_gene_available=is_gene_available,
                            is_trait_available=is_trait_available)

# 4. Extract clinical features since trait_row is not None
clinical_df = geo_select_clinical_features(clinical_df=clinical_data,
                                         trait=trait,
                                         trait_row=trait_row,
                                         convert_trait=convert_trait,
                                         age_row=age_row,
                                         convert_age=convert_age,
                                         gender_row=gender_row,
                                         convert_gender=convert_gender)

print("\nPreview of processed clinical data:")
print(preview_df(clinical_df))

# Save clinical data
clinical_df.to_csv(out_clinical_data_file)
# Extract gene expression data from matrix file
genetic_data = get_genetic_data(matrix_file_path)

# Print first 20 row IDs 
print("First 20 gene/probe IDs:")
print(list(genetic_data.index[:20]))
# The gene identifiers have the prefix "ILMN_" which indicates they are Illumina probe IDs
# These are not standard human gene symbols and will need to be mapped
requires_gene_mapping = True
# Extract gene annotation from SOFT file 
gene_annotation = get_gene_annotation(soft_file_path)

# Preview annotation structure
preview = preview_df(gene_annotation)
print("Gene annotation preview:")
print(preview)
# 1. Identify relevant columns for mapping
# 'ID' column in annotation matches the ILMN_* probe IDs in expression data
# 'Symbol' column contains the human gene symbols
probe_col = 'ID'
gene_col = 'Symbol'

# 2. Get mapping dataframe
mapping_df = get_gene_mapping(gene_annotation, probe_col, gene_col)

# 3. Apply mapping to convert probe data to gene expression data 
gene_data = apply_gene_mapping(genetic_data, mapping_df)

# Preview first few rows of converted data
print("\nPreview of gene expression data after mapping:")
print(preview_df(gene_data))
# 1. Normalize gene symbols in gene expression data
gene_data = normalize_gene_symbols_in_index(gene_data)
os.makedirs(os.path.dirname(out_gene_data_file), exist_ok=True)
gene_data.to_csv(out_gene_data_file)
print("\nGene data shape (normalized gene-level):", gene_data.shape) 

# Load clinical data previously processed
selected_clinical_df = pd.read_csv(out_clinical_data_file, index_col=0)
print("\nClinical data shape:", selected_clinical_df.shape)

# 2. Link clinical and genetic data using normalized gene-level data
linked_data = geo_link_clinical_genetic_data(selected_clinical_df, gene_data)
print("\nLinked data shape:", linked_data.shape)

# 3. Handle missing values systematically  
if trait in linked_data.columns:
    linked_data = handle_missing_values(linked_data, trait)

    # 4. Check for bias in trait and demographic features
    trait_biased, linked_data = judge_and_remove_biased_features(linked_data, trait)

    # 5. Final validation and information saving
    note = "Data was successfully preprocessed from probe-level to gene-level expression using gene symbol normalization with NCBI Gene database."
    is_usable = validate_and_save_cohort_info(
        is_final=True,
        cohort=cohort, 
        info_path=json_path,
        is_gene_available=True,
        is_trait_available=True,
        is_biased=trait_biased,
        df=linked_data,
        note=note
    )

    # 6. Save linked data only if usable and not biased
    if is_usable and not trait_biased:
        os.makedirs(os.path.dirname(out_data_file), exist_ok=True)
        linked_data.to_csv(out_data_file)