🚀 Overview

This project is based on our paper "Complex structure-free compound-protein interaction prediction for mitigating activity cliff-induced discrepancies and integrated bioactivity learning". GGAP-CPI stands for protein Graph and ligand Graph network with Attention Pooling for Compound-Protein Interaction prediction.

📖 Table of Contents

• 🎯 Introduction
• 🗄️ Dataset
• 📦 Dependencies
• 🏋️‍♂️ Model Training
• 🔄 Model Fine‑tuning
• 🔍 Model Inference
• 📊 Benchmark Results
• 🛠️ Use Your Own Data
• 📢 Citation

🎯 Introduction

Protein-ligand binding affinity assessment plays a pivotal role in virtual drug screening, yet conventional data-driven approaches rely heavily on limited protein-ligand crystal structures. Structure-free compound-protein interaction (CPI) methods have emerged as competitive alternatives, leveraging extensive bioactivity data to serve as more robust scoring functions. However, these methods often overlook two critical challenges that affect data efficiency and modeling accuracy: the heterogeneity of bioactivity data due to differences in bioassay measurements, and the presence of activity cliffs (ACs)—small chemical modifications that lead to significant changes in bioactivity, which have not been thoroughly investigated in CPI modeling.

To address these challenges, we present CPI2M, a large-scale CPI benchmark dataset containing approximately 2 million bioactivity endpoints across four activity types (Ki, Kd, EC50, and IC50) with AC annotations. Moreover, we developed GGAP-CPI-IntEns, a complex structure-free deep learning model trained by integrated bioactivity learning and designed to mitigate the impact of ACs on CPI prediction through advanced protein representation modelling and integrated bioactivity learning.

Our comprehensive evaluation demonstrates that GGAP-CPI-IntEns outperforms 12 target-specific and 7 general CPI baselines across four key scenarios (general CPI prediction, rare protein prediction, transfer learning, and virtual screening) on seven benchmarks (CPI2M, MoleculeACE, CASF-2016, MerckFEP, DUD-E, DEKOIS-v2, and LIT-PCBA). Furthermore, GGAPCPI-IntEns not only delivers stable predictions by distinguishing bioactivity differences between ACs and non-ACs, but also enriches binding pocket residues and interactions, underscoring its applicability to real-world binding affinity assessments and virtual drug screening.

🗄️ Dataset

Bioactivity Dataset Summary

Dataset	Activity Type	Num.	Num. Mol.	Num. Prot.	Avg. Bioactivity	Std. Bioactivity	% AC
CPI2M-main (train, internal validation)	Ki	341,244	124,345	418	6.50	1.42	25.39
	Kd	4,337	3,212	21	6.90	1.60	34.03
	EC50	88,302	61,095	178	5.80	1.56	25.08
	IC50	751,941	419,985	1115	6.15	1.47	30.60
CPI2M-few (external validation)	Ki	65,529	41,365	2373	6.20	1.60	-
	Kd	55,017	14,667	1564	5.79	1.35	-
	EC50	42,301	28,818	1506	6.00	1.48	-
	IC50	148,929	94,883	4562	5.69	1.43	-

We also incorporate MoleculeACE for activity cliff estimation, CASF-2016, MerckFEP, DUD-E, DEKOIS-2, and LIT-PCBA for virtual screening estimation.

Description of CPI2M

• Source: EquiVS (ChEMBL29, BindingDB, PubChem, Probe&Drugs, IUPHAR/BPS), and Papyrus (ChEMBL30, EXCAPE, literature)
• Structure: CPI2M-main for model training and internal evaluating, CPI2M-few for external evaluating.
• Preprocessing: Including multistep filtering and duplicate cleaning for activity value, unit, ligand, and protein data.

The access of full CPI2M dataset is available at Zenodo: CPI2M.

📦 Dependencies

• torch==2.4.1+cu121
• torch-geometric==2.6.1
• torch-scatter==2.1.2+pt24cu121
• torch-sparse==0.6.18+pt24cu121
• torch-spline-conv==1.2.2+pt24cu121
• fair-esm==2.0.0
• chemprop==1.6.1
• DeepPurpose==0.1.5
• MoleculeACE==2.0.1
• graphein==1.7.5
• rdkit==2023.9.1
• MolVS==0.1.1
• biopython==1.81
• scikit-learn==1.3.2
• networkx==3.1
• numpy
• pandas
• yaml

Model

🏋️‍♂️ Model Training

Please run the following command for model training:

sh run_bash/run_CPI.sh GGAP_CPI {DATA_NAME} train {SEED}

parameters include: 1. training dataset; 2. mode (e.g., train); 3. random seed.

🔄 Model Finetuning

To use the pretrained GGAP-CPI-IntEns model (ensemble of 10 GGAP-CPI models) for finetuing on your specific dataset, please run the following command:

for seed in $(seq 0 9); do
model_path=GGAP_CPI_IntEns_${seed}
sh run_bash/run_CPI.sh GGAP_CPI {DATA_NAME} finetune {SEED} ${model_path}
done

🔍 Model Inference

Taking "kd.csv" in data folder for example, please run the following command for inferencing:

example_data=kd
example_model_path=GGAP_CPI_IntEns_0
example_seed=0
sh run_bash/run_CPI.sh GGAP_CPI ${example_data} inference ${example_seed} ${example_model_path}

📊 Benchmark Results

The performances of GGAP-CPI and 19 baseline methods are evaluated on CPI2M-main, CPI2M-few, MoleculeACE, CASF-2016, MerckFEP, DUD-E, DEKOIS-2, and LIT-PCBA datasets. For your convience, we add the benchmarking result files for each of them in "benchmark_result" folder.

🛠️ Use Your Own Data

To train GGAP-CPI from scratch on your own .CSV data, which should at least include columns ['smiles', 'Uniprot_id', 'label']. Note that the 'Uniprot_id' can be either the protein UniProt ID that can be accessed from Alphafold2 database or the protein PDB file name that have been stored in the "data/PDB" folder. For raw PDB file, we will automatically extract the first chain as the protein structure for training and testing. Please run the following commands:

# preprocess data
python process_data.py --dataset {DATA} --task {CPI or QSAR} --split {random or ac} --train_ratio {RATIO} --seed {SEED}
# train
sh run_bash/run_CPI.sh {MODEL} {DATA_NAME} train {SEED}
# optional: finetune, inference, ...

📢 Citation

@article{GGAP_CPI,
   author = {Gu, Yaowen and Xia, Song and Ouyang, Qi and Zhang, Yingkai},
   title = {Complex structure-free compound-protein interaction prediction for mitigating activity cliff-induced discrepancies and integrated bioactivity learning},
   DOI = {10.26434/chemrxiv-2025-96p6b},
   year = {2025},
   type = {Journal Article}
}