Liu-Hy's picture
Add files using upload-large-folder tool
0a8b3ff verified
# Path Configuration
from tools.preprocess import *
# Processing context
trait = "Bladder_Cancer"
cohort = "GSE222073"
# Input paths
in_trait_dir = "../DATA/GEO/Bladder_Cancer"
in_cohort_dir = "../DATA/GEO/Bladder_Cancer/GSE222073"
# Output paths
out_data_file = "./output/preprocess/3/Bladder_Cancer/GSE222073.csv"
out_gene_data_file = "./output/preprocess/3/Bladder_Cancer/gene_data/GSE222073.csv"
out_clinical_data_file = "./output/preprocess/3/Bladder_Cancer/clinical_data/GSE222073.csv"
json_path = "./output/preprocess/3/Bladder_Cancer/cohort_info.json"
# Get file paths
soft_file, matrix_file = geo_get_relevant_filepaths(in_cohort_dir)
# Extract background info and clinical data
background_info, clinical_data = get_background_and_clinical_data(matrix_file)
# Get unique values per clinical feature
sample_characteristics = get_unique_values_by_row(clinical_data)
# Print background info
print("Dataset Background Information:")
print(f"{background_info}\n")
# Print sample characteristics
print("Sample Characteristics:")
for feature, values in sample_characteristics.items():
print(f"Feature: {feature}")
print(f"Values: {values}\n")
# 1. Gene Expression Data Availability
# Yes, this appears to be a gene expression dataset based on mentions of RNA subtypes and labeling kits
is_gene_available = True
# 2.1 Data Availability
# Trait: Available from various metastasis indicators (rm-* fields)
# Will use rm-bone as representative of metastasis
trait_row = 11
# Age and Gender not available in sample characteristics
age_row = None
gender_row = None
# 2.2 Data Type Conversion Functions
def convert_trait(value: str) -> int:
"""Convert bone metastasis status to binary (0: no, 1: yes)"""
if not isinstance(value, str):
return None
value = value.lower()
if 'rm-bone:' not in value:
return None
value = value.split('rm-bone:')[1].strip()
if value == 'yes':
return 1
elif value == 'no':
return 0
return None
def convert_age(value: str) -> float:
"""Placeholder for age conversion"""
return None
def convert_gender(value: str) -> int:
"""Placeholder for gender conversion"""
return None
# 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. Clinical Feature Extraction
if trait_row is not None:
selected_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)
# Preview the data
preview = preview_df(selected_clinical_df)
print("Preview of clinical data:")
print(preview)
# Save clinical data
selected_clinical_df.to_csv(out_clinical_data_file)
# Extract gene expression data from matrix file
gene_data = get_genetic_data(matrix_file)
# Print first 20 row IDs and shape of data to help debug
print("Shape of gene expression data:", gene_data.shape)
print("\nFirst few rows of data:")
print(gene_data.head())
print("\nFirst 20 gene/probe identifiers:")
print(gene_data.index[:20])
# Inspect a snippet of raw file to verify identifier format
import gzip
with gzip.open(matrix_file, 'rt', encoding='utf-8') as f:
lines = []
for i, line in enumerate(f):
if "!series_matrix_table_begin" in line:
# Get the next 5 lines after the marker
for _ in range(5):
lines.append(next(f).strip())
break
print("\nFirst few lines after matrix marker in raw file:")
for line in lines:
print(line)
# Looking at the gene identifiers, we see a mix of symbols like 'A2M', 'A4GALT' which are human gene symbols,
# and some numbered identifiers like '1-Mar', '2-Mar' etc.
# The presence of 'Mar' suggests these might be month-related probe identifiers that need mapping.
requires_gene_mapping = True
# Get file paths using library function
soft_file, matrix_file = geo_get_relevant_filepaths(in_cohort_dir)
# First locate the platform table section in SOFT file
import gzip
table_begin = None
table_end = None
with gzip.open(soft_file, 'rt', encoding='utf-8') as f:
for i, line in enumerate(f):
if '!platform_table_begin' in line.lower():
table_begin = i
elif '!platform_table_end' in line.lower() and table_begin is not None:
table_end = i
break
# Read annotation data between markers using skiprows and nrows
import pandas as pd
if table_begin is not None and table_end is not None:
gene_annotation = pd.read_csv(soft_file, compression='gzip', skiprows=table_begin+1,
nrows=table_end-table_begin-1, sep='\t')
else:
# Fallback to original method if markers not found
gene_annotation = get_gene_annotation(soft_file)
# Preview gene annotation data
print("Gene annotation preview:")
print(preview_df(gene_annotation))
# Display all column names
print("\nAll column names in annotation data:")
print(gene_annotation.columns.tolist())
# In this dataset, the probe ID "A2M" in the gene expression data should be matched to the gene symbol in 'ORF' column
gene_mapping = get_gene_mapping(annotation=gene_annotation, prob_col='ID', gene_col='ORF')
# Convert probe-level measurements to gene expression data
gene_data = apply_gene_mapping(expression_df=gene_data, mapping_df=gene_mapping)
# Save the gene expression data
gene_data.to_csv(out_gene_data_file)
# Preview results
print("Shape of mapped gene expression data:", gene_data.shape)
print("\nFirst few rows of mapped data:")
print(gene_data.head())
# 1. Normalize gene symbols and save normalized gene data
# Remove "-mRNA" suffix from gene symbols before normalization
gene_data.index = gene_data.index.str.replace('-mRNA', '')
gene_data = normalize_gene_symbols_in_index(gene_data)
gene_data.to_csv(out_gene_data_file)
# 2. Link clinical and genetic data and trait
# First get selected clinical features using the extraction function from previous step
selected_clinical = 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
)
# Debug data structures before linking
print("\nPre-linking data shapes:")
print("Clinical data shape:", selected_clinical.shape)
print("Gene data shape:", gene_data.shape)
print("\nClinical data preview:")
print(selected_clinical.head())
# Transpose gene data to match clinical data orientation
gene_data_t = gene_data.T
linked_data = pd.concat([selected_clinical.T, gene_data_t], axis=1)
# 3. Handle missing values systematically
linked_data = handle_missing_values(linked_data, trait)
# 4. Check for biased features and remove them if needed
is_biased, linked_data = judge_and_remove_biased_features(linked_data, trait)
# 5. Validate data quality and save metadata
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=is_biased,
df=linked_data,
note="Gene expression data from pancreatic cancer study. All samples are cancer cases (no controls)."
)
# 6. Save linked data if usable
if is_usable:
linked_data.to_csv(out_data_file)