data/generate_h5ad.py

import warnings
warnings.filterwarnings("ignore")
import os
import logging
import random
import torch
import sys

# import hdf5plugin
import numpy as np
import anndata as ad
from scipy.sparse import csr_matrix
import scanpy as sc
import matplotlib.pyplot as plt
import json
import pandas as pd
import h5py
import numpy as np
import scanpy as sc
import numpy as np
import pandas as pd



destination_file = '/egr/research-dselab/pujing1/foundation_bulk/data/human_gene_v2.5.h5'
metadata_train = pd.read_csv("/egr/research-dselab/pujing1/foundation_bulk/data/GEO_HUMAN_500K_training.csv")
metadata_test = pd.read_csv("/egr/research-dselab/pujing1/foundation_bulk/data/GEO_HUMAN_500K_test.csv")
metadata_test_2024 = pd.read_csv("/egr/research-dselab/pujing1/foundation_bulk/data/GEO_HUMAN_500K_test_2024.csv")
metadata = pd.concat([metadata_train, metadata_test, metadata_test_2024], ignore_index=True)
metadata['GSM_ID'] = [bytes(sample, 'utf-8') for sample in metadata['GSM_ID']]
metadata_train['GSM_ID'] = [bytes(sample, 'utf-8') for sample in metadata_train['GSM_ID']]
metadata_test['GSM_ID'] = [bytes(sample, 'utf-8') for sample in metadata_test['GSM_ID']]
metadata_test_2024['GSM_ID'] = [bytes(sample, 'utf-8') for sample in metadata_test_2024['GSM_ID']]


with h5py.File(destination_file, 'r') as f:
    num_samples = len(f['meta/samples/geo_accession'][:])
    expression_dataset = f['data/expression']
    total_genes, total_samples = expression_dataset.shape

    batch_size = 1000 
    expression_chunks = []
    missing_gene_ranges = [(56000, 57000), (57000, 58000)]

    total_genes = 67186
    gene_mask = np.ones(total_genes, dtype=bool)
    for start, end in missing_gene_ranges:
        gene_mask[start:end] = False


    for start_row in range(0, total_genes, batch_size):
        end_row = min(start_row + batch_size, total_genes)
        try:
            chunk = expression_dataset[start_row:end_row, :]
            expression_chunks.append(chunk)
            print(f"Successfully read {start_row}:{end_row} ({end_row-start_row} genes)")
            
        except Exception as e:
            print(f"Error reading genes {start_row}:{end_row}: {e}")
            continue
    
    if expression_chunks:
        print("Chunks read successfully, starting to merge...")
        expression = np.vstack(expression_chunks)
        print(f"Merging completed, final shape: {expression.shape}")
    ensembl_id = f['meta/genes/ensembl_gene'][gene_mask]
    symbol = f['meta/genes/symbol'][gene_mask]
    geo_accession = f['meta/samples/geo_accession'][:]
    sample = f['meta/samples/sample'][:]
    series_id = f['meta/samples/series_id'][:]
    adata = ad.AnnData(X=expression.T, obs={'geo_accession': geo_accession, 'sample': sample, 'series_id': series_id}, 
    var={'ensembl_id': ensembl_id, 'symbol': symbol})
    adata.var['ensembl_id'] = adata.var['ensembl_id'].apply(lambda x: x.decode('utf-8'))
    adata.var['ensembl.gene'] = adata.var['ensembl_id']
    adata.var.index = adata.var['ensembl.gene']
    adata.obs['geo_accession'] = adata.obs['geo_accession'].astype('category')
    adata.obs['sample'] = adata.obs['sample'].astype('category')
    adata.obs['series_id'] = adata.obs['series_id'].astype('category')
    metadata_batch = metadata[metadata['GSM_ID'].isin(adata.obs['geo_accession'])]
    duplicates = metadata_batch[metadata_batch.duplicated(subset='GSM_ID', keep=False)]
    print(f"Number of duplicate GSM_IDs: {len(duplicates)}")
    metadata_batch = metadata_batch.drop_duplicates(subset='GSM_ID', keep='first')

    merged_gender = adata.obs.merge(metadata_batch[['GSM_ID', 'Gender']], 
                        left_on='geo_accession', 
                        right_on='GSM_ID', 
                        how='left')
    
    adata.obs['gender'] = merged_gender['Gender'].values

    merged_organ = adata.obs.merge(metadata_batch[['GSM_ID', 'Organ']], 
                        left_on='geo_accession', 
                        right_on='GSM_ID', 
                        how='left')
    adata.obs['organ_system'] = merged_organ['Organ'].values

    merged_disease = adata.obs.merge(metadata_batch[['GSM_ID', 'Disease']], 
                        left_on='geo_accession', 
                        right_on='GSM_ID', 
                        how='left')
    adata.obs['disease'] = merged_disease['Disease'].values
    adata.obs['disease'] = adata.obs['disease'].replace('Healthy', 'Normal')

    merged_age = adata.obs.merge(metadata_batch[['GSM_ID', 'Age_Group']], 
                        left_on='geo_accession', 
                        right_on='GSM_ID', 
                        how='left')
    adata.obs['age'] = merged_age['Age_Group'].values

    merged_experimental_setting = adata.obs.merge(metadata_batch[['GSM_ID', 'Exp_Setting']], 
                        left_on='geo_accession', 
                        right_on='GSM_ID', 
                        how='left')

    adata.obs['Experimental_Setting'] = merged_experimental_setting['Exp_Setting'].values

    disease_map = pd.read_excel(
        '/egr/research-dselab/pujing1/foundation_bulk/data/Disease_groups_check_reviewed_with_reasoning_UPDATED_05112026.xlsx',
        sheet_name='Sheet1',
        usecols=['Disease_Post', 'Broad_Disease_Category']
    ).drop_duplicates(subset='Disease_Post').set_index('Disease_Post')['Broad_Disease_Category']
    adata.obs['Broad_Disease_Category'] = adata.obs['disease'].map(disease_map)


adata_train = adata[adata.obs['geo_accession'].isin(metadata_train['GSM_ID'])]
adata_train.write('/egr/research-dselab/pujing1/foundation_bulk/data/human_gene_v2.5_train_nhmerged.h5ad')
adata_test = adata[adata.obs['geo_accession'].isin(metadata_test['GSM_ID'])]
adata_test.write('/egr/research-dselab/pujing1/foundation_bulk/data/human_gene_v2.5_test_nhmerged.h5ad')
adata_test_2024 = adata[adata.obs['geo_accession'].isin(metadata_test_2024['GSM_ID'])]
adata_test_2024.write('/egr/research-dselab/pujing1/foundation_bulk/data/human_gene_v2.5_test_2024_nhmerged.h5ad')