CheapNet: Cross-attention on Hierarchical representations for Efficient protein-ligand binding Affinity Prediction

Hyukjun Lim, Sun Kim, Sangseon Lee†
Seoul National University, Inha University
ICLR 2025
† Indicates Corresponding Author
Paper OpenReview Code Poster
Representative Case study of CheapNet

Visualization of CheapNet interpretation and cross-attention map of a protein-ligand complex.
(a) The high-attended pairs of protein and ligand atoms are highlighted in the red box.
(b) The high-attented regions of cross-attention map marked in yellow within the red box.

Abstract

Accurately predicting protein-ligand binding affinity is a critical challenge in drug discovery, crucial for understanding drug efficacy. While existing models typically rely on atom-level interactions, they often fail to capture the complex, higher-order interactions, resulting in noise and computational inefficiency. Transitioning to modeling these interactions at the cluster level is challenging because it is difficult to determine which atoms form meaningful clusters that drive the protein-ligand interactions. To address this, we propose CheapNet, a novel interaction-based model that integrates atom-level representations with hierarchical cluster-level interactions through a cross-attention mechanism. By employing soft clustering of atom-level embeddings, CheapNet efficiently captures essential higher-order molecular representations crucial for accurate binding predictions. Extensive evaluations demonstrate that CheapNet not only achieves state-of-the-art performance across multiple binding affinity prediction tasks but also maintains prediction accuracy with reasonable computational efficiency.

Method

Graph Encoder: First encode the protein-ligand complex as a graph, where nodes represent atoms and edges represent bonds. Then learn atom-level embeddings using a graph neural networks (GNNs).
Soft clustering: Leverage soft clustering to cluster atom-level embeddings into cluster-level representations. This allows us to capture higher-order interactions between atoms.
Cross-Attention Mechanism: Apply cross-attention mechanism between protein and ligand clusters to capture key interactions. This enables the model to focus on the most informative regions of the protein-ligand complex.

Overview of CheapNet

Architecture of CheapNet for protein-ligand binding affinity prediction.
(a) Graph Encoder, (b) Soft Clustering based on Embeddings, and (c) Cross-Attention Mechanism.

Results

CheapNet achieves superior performance in protein-ligand binding affinity prediction tasks, including cross-dataset evaluation, diverse protein evaluation, and ligand efficacy prediction. It also offers enhanced interpretability by highlighting key protein-ligand interactions. Additionally, CheapNet maintains high prediction accuracy with reasonable computational efficiency, making it more promising.

BibTeX

@inproceedings{
  lim2025cheapnet,
  title={CheapNet: Cross-attention on Hierarchical representations for Efficient protein-ligand binding Affinity Prediction},
  author={Hyukjun Lim and Sun Kim and Sangseon Lee},
  booktitle={The Thirteenth International Conference on Learning Representations},
  year={2025},
  url={https://openreview.net/forum?id=A1HhtITVEi}
}