Zhenghang04/DeepSeek_DrugSmiles

Small Molecule Datasets (CMAP, Tahoe, CIGS)

Dataset Introduction

This dataset hosts standardized small molecule chemical data integrated with DeepSeek ecological adaptation, supporting one-click download and quick loading for AI molecular modeling, computational chemistry and drug property prediction tasks. Molecular structures are uniformly stored as canonical SMILES strings, sourced from CMAP, Tahoe and CIGS public libraries. All raw data has been cleaned, unified and converted into universal CSV & HDF5 formats, perfectly compatible with DeepSeek model fine-tuning, molecular feature extraction and cheminformatics downstream experiments.

Language

Dataset core files adopt English standard specification, with bilingual auxiliary documents, convenient for domestic and foreign researchers to use in DeepSeek related research.

Repository Directory Structure

.
├── data/
│   ├── raw/                  # Original unprocessed raw data package
│   ├── processed/            # Preprocessed HDF5 intermediate data
│   │   ├── CIGS/             # CIGS cell sample classified data
│   │   ├── CIGS_processed_data_id.h5
│   │   └── tahoe_processed_data_id.h5
│   └── smiles/               # Standard SMILES CSV dataset (Direct download available)
│       ├── CMAP canonical_smiles.csv
│       ├── CIGS canonical_smiles.csv
│       └── tahoe canonical_smiles.csv
├── docs/                     # Dataset usage docs matching DeepSeek training
│   ├── content.pdf
│   └── dataset_parsing.md
├── scripts/                  # Format conversion script adapting DeepSeek input
│   └── convert_h5_to_csv.py
└── README.md

Data Column Description

All downloadable CSV files under data/smiles contain two standard fields:

• index: Unique serial ID of each small molecule
• canonical_smiles: Standard molecular structure SMILES sequence, directly feedable into DeepSeek molecular model

Quick Download & Loading Guide

1. Direct Local File Download

Download three SMILES CSV files in the data/smiles folder directly, no complicated decompression processing required, ready for immediate use.

2. Load with Pandas (Suitable for DeepSeek data preprocessing)

import pandas as pd

# Load three molecular datasets separately
cmap_df = pd.read_csv("data/smiles/CMAP canonical_smiles.csv")
tahoe_df = pd.read_csv("data/smiles/tahoe canonical_smiles.csv")
cigs_df = pd.read_csv("data/smiles/CIGS canonical_smiles.csv")

# Quick preview, directly used for DeepSeek feature construction
print(cmap_df.head())

3. Load via Hugging Face Dataset Library (One-click import for DeepSeek training)

from datasets import load_dataset

# Batch import all datasets, adaptive DeepSeek model input format
dataset = load_dataset("csv", data_files={
    "cmap": "data/smiles/CMAP canonical_smiles.csv",
    "tahoe": "data/smiles/tahoe canonical_smiles.csv",
    "cigs": "data/smiles/CIGS canonical_smiles.csv"
})
print(dataset)

Data Processing Adaptation for DeepSeek

Original Tahoe and CIGS data are stored in HDF5 format. The built-in conversion script can rapidly convert data into SMILES standard format consistent with CMAP, fully matching the input specification of DeepSeek large model molecular tasks:

• Tahoe: Directly extract standard SMILES from HDF5 file
• CIGS: Complete index-molecular structure mapping parsing, unified model training dimension

Citation

When utilizing this dataset for DeepSeek model research, paper publication and experimental development, please cite the original CMAP, Tahoe and CIGS data sources.