Liu-Hy's picture
Add files using upload-large-folder tool
75faa94 verified
# Path Configuration
from tools.preprocess import *
# Processing context
trait = "Eczema"
cohort = "GSE150797"
# Input paths
in_trait_dir = "../DATA/GEO/Eczema"
in_cohort_dir = "../DATA/GEO/Eczema/GSE150797"
# Output paths
out_data_file = "./output/preprocess/3/Eczema/GSE150797.csv"
out_gene_data_file = "./output/preprocess/3/Eczema/gene_data/GSE150797.csv"
out_clinical_data_file = "./output/preprocess/3/Eczema/clinical_data/GSE150797.csv"
json_path = "./output/preprocess/3/Eczema/cohort_info.json"
# Get paths to the SOFT and matrix files
soft_file, matrix_file = geo_get_relevant_filepaths(in_cohort_dir)
# Get background info and clinical data from matrix file
background_info, clinical_data = get_background_and_clinical_data(matrix_file)
# Get unique values for each feature (row) in clinical data
unique_values_dict = get_unique_values_by_row(clinical_data)
# Print background info
print("=== Dataset Background Information ===")
print(background_info)
print("\n=== Sample Characteristics ===")
print(json.dumps(unique_values_dict, indent=2))
# 1. Gene Expression Data Availability
# Yes, microarray data (Affymetrix) is available according to background info
is_gene_available = True
# 2. Variable Availability and Type Conversion
trait_row = 2 # treatment status indicates disease severity
age_row = None # age not provided
gender_row = 1 # gender is available
def convert_trait(value: str) -> Optional[int]:
"""Convert treatment status to disease severity binary"""
if not isinstance(value, str):
return None
value = value.split(": ")[-1].lower().strip()
if "untreated" in value:
return 1 # Active disease
elif "treated" in value or "nb-uvb" in value:
return 0 # Treated/controlled disease
return None
def convert_gender(value: str) -> Optional[int]:
"""Convert gender to binary (0=female, 1=male)"""
if not isinstance(value, str):
return None
value = value.split(": ")[-1].lower().strip()
if value == "female":
return 0
elif value == "male":
return 1
return None
# 3. Save Initial Metadata
is_usable = validate_and_save_cohort_info(
is_final=False,
cohort=cohort,
info_path=json_path,
is_gene_available=is_gene_available,
is_trait_available=trait_row is not None
)
# 4. Extract Clinical Features
if 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,
gender_row=gender_row,
convert_gender=convert_gender
)
# Preview the extracted features
preview = preview_df(clinical_df)
# Save clinical data
os.makedirs(os.path.dirname(out_clinical_data_file), exist_ok=True)
clinical_df.to_csv(out_clinical_data_file)
# Extract gene expression data from matrix file
genetic_df = get_genetic_data(matrix_file)
# Print DataFrame shape and first 20 row IDs
print("DataFrame shape:", genetic_df.shape)
print("\nFirst 20 row IDs:")
print(genetic_df.index[:20])
print("\nPreview of first few rows and columns:")
print(genetic_df.head().iloc[:, :5])
# These identifiers appear to be Affymetrix probeset IDs (TC identifiers followed by .hg.1)
# They need to be mapped to standard human gene symbols
requires_gene_mapping = True
# Extract gene annotation data, excluding control probe lines
gene_metadata = get_gene_annotation(soft_file)
# Function to extract gene symbol from the text description
def extract_gene_symbol(text):
if not isinstance(text, str):
return None
# Look for RefSeq gene name
refseq_match = re.search(r'RefSeq // Homo sapiens ([\w\-]+) \(', text)
if refseq_match:
return refseq_match.group(1)
# Look for ENSEMBL gene name
ensembl_match = re.search(r'ENSEMBL // ([\w\-]+) \[', text)
if ensembl_match:
return ensembl_match.group(1)
# Look for gene name with HGNC Symbol
hgnc_match = re.search(r'\[Source:HGNC Symbol;Acc:HGNC:\d+\] // ([\w\-]+)', text)
if hgnc_match:
return hgnc_match.group(1)
return None
# Add gene symbol column and create mapping dataframe
gene_metadata['Gene_Symbol'] = gene_metadata['SPOT_ID.1'].apply(extract_gene_symbol)
gene_mapping = gene_metadata[['ID', 'Gene_Symbol']].dropna()
gene_mapping = gene_mapping.rename(columns={'Gene_Symbol': 'Gene'})
# Preview filtered annotation data
print("Column names:")
print(gene_metadata.columns)
print("\nPreview of gene mapping data:")
print(preview_df(gene_mapping))
# Print mapping statistics
n_total = len(gene_metadata)
n_mapped = len(gene_mapping)
print(f"\nTotal probes: {n_total}")
print(f"Probes mapped to gene symbols: {n_mapped} ({n_mapped/n_total:.1%})")
# Based on the preview, refine the gene mapping strategy
# Extract probes from SOFT file annotations with better gene symbol extraction
gene_metadata = get_gene_annotation(soft_file)
def extract_gene_symbol(text):
if not isinstance(text, str):
return None
# Look for RefSeq gene name
refseq_match = re.search(r'RefSeq.*?//(.*?)\[', text)
if refseq_match:
symbol = refseq_match.group(1).strip()
if not symbol.startswith(('LOC', 'LINC')):
return symbol
# Look for HGNC Symbol
hgnc_match = re.search(r'HGNC.*?//(.*?)\[', text)
if hgnc_match:
return hgnc_match.group(1).strip()
return None
# Create mapping dataframe with probe IDs and gene symbols
gene_metadata['Gene'] = gene_metadata['SPOT_ID.1'].apply(extract_gene_symbol)
mapping_df = gene_metadata[['ID', 'Gene']].dropna()
# Apply gene mapping to convert probe-level data to gene-level expression
gene_data = apply_gene_mapping(genetic_df, mapping_df)
# Preview the results
print("Gene mapping statistics:")
print(f"Total probes: {len(genetic_df)}")
print(f"Probes mapped to genes: {len(mapping_df)}")
print(f"Final unique genes: {len(gene_data)}")
print("\nPreview of gene expression data:")
print(gene_data.head())
# 1. Normalize gene symbols and save
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)
# 2. Link clinical and genetic data
clinical_df = pd.read_csv(out_clinical_data_file, index_col=0)
linked_data = geo_link_clinical_genetic_data(clinical_df, gene_data)
# 3. Handle missing values
linked_data = handle_missing_values(linked_data, trait)
# 4. Check for biased features
trait_biased, linked_data = judge_and_remove_biased_features(linked_data, trait)
# 5. Final validation and metadata saving
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="Study comparing gene expression in healthy vs DMD myoblasts and myotubes, including immortalized cell lines"
)
# 6. Save linked data if usable
if is_usable:
os.makedirs(os.path.dirname(out_data_file), exist_ok=True)
linked_data.to_csv(out_data_file)