GDEGAN: Gaussian Dynamic Equivariant Graph Attention Network for Ligand Binding Site Prediction

Animesh; Plaban Kumar Bhowmick; Pralay Mitra

cs.LGMar 20, 2026

GDEGAN: Gaussian Dynamic Equivariant Graph Attention Network for Ligand Binding Site Prediction

Authors: Animesh, Plaban Kumar Bhowmick, Pralay Mitra

Organizations: Department of Artificial Intelligence, Indian Institute of Technology, Kharagpur, India · Department of Computer Science, Indian Institute of Technology, Kharagpur, India

Abstract

Accurate prediction of binding sites of a given protein, to which ligands can bind, is a critical step in structure-based computational drug discovery. Recently, Equivariant Graph Neural Networks (GNNs) have emerged as a powerful paradigm for binding site identification methods due to the large-scale availability of 3D structures of proteins via protein databases and AlphaFold predictions. The state-of-the-art equivariant GNN methods implement dot product attention, disregarding the variation in the chemical and geometric properties of the neighboring residues. To capture this variation, we propose GDEGAN (Gaussian Dynamic Equivariant Graph Attention Network), which replaces dot-product attention with adaptive kernels that recognize binding sites. The proposed attention mechanism captures variation in neighboring residues using statistics of their characteristic local feature distributions. Our mechanism dynamically computes neighborhood statistics at each layer, using local variance as an adaptive bandwidth parameter with learnable per-head temperatures, enabling each protein region to determine its own context-specific importance. GDEGAN outperforms existing methods with relative improvements of 37-66% in DCC and 7-19% DCA success rates across COACH420, HOLO4k, and PDBBind2020 datasets. These advances have direct application in accelerating protein-ligand docking by identifying potential binding sites for therapeutic target identification.

Figures & tables

Figure 1: GDEGAN architecture for protein ligand binding site identification. Left: Overview of the GDEGAN framework showing the integration of protein-specific ESM-2 embeddings with geometric processing through $L$ layers of Gaussian Dynamic Attention. Right: Detailed view of the Gaussian Dynamic Attention (GDA) module. In this $\oplus$ , $\cdot$ and $\circ$ denotes addition, dot product and element-wise product respectively. HTR is inherited from GotenNet ( Aykent and Xia, 2025 ) . Soft. stands for Softmax, Agg. for Aggregation.

Methods	Param (M)	Failure rate $\downarrow$	COACH420		HOLO4K k		PDBbind2020
Methods	Param (M)	Failure rate $\downarrow$	DCC $\uparrow$	DCA $\uparrow$	DCC $\uparrow$	DCA $\uparrow$	DCC $\uparrow$	DCA $\uparrow$
Fpocket b	$\backslash$	0.000	0.228	0.444	0.192	0.457	0.253	0.371
P2rank b	$\backslash$	0.000	0.366	0.628	0.314	0.621	0.503	0.677
DeepSite b	1.00	$\backslash$	$\backslash$	0.564	$\backslash$	0.456	$\backslash$	$\backslash$
Kalasanty b	70.6	0.120	0.335	0.636	0.244	0.515	0.416	0.625
DeepSurf b	33.1	0.054	0.386	0.658	0.289	0.635	0.510	0.708

Table 1: Experimental results of baseline models and our framework measured by DCC and DCA success rates a . The table presents comparative results across three benchmark datasets. Bold values indicate the best performance in each metric.

Methods	EQ c	ADL d	ESM e	COACH420		HOLO4K		PDBbind2020
Methods	EQ c	ADL d	ESM e	DCC $\uparrow$	DCA $\uparrow$	DCC $\uparrow$	DCA $\uparrow$	DCC $\uparrow$	DCA $\uparrow$
GotenNet	$E(3)$	No	No	0.454(0.007)	0.624(0.014)	0.464(0.001)	0.691(0.005)	0.553(0.008)	0.705(0.007)
GotenNet+ADL	$E(3)$	Yes	No	0.485(0.004)	0.642(0.011)	0.468(0.004)	0.732(0.004)	0.592(0.010)	0.748(0.003)
GotenNet+ESM	$SE(3)$	No	Yes	0.543(0.008)	0.693(0.006)	0.520(0.011)	0.753(0.004)	0.637(0.005)	0.760(0.004)
GotenNet(full)	$SE(3)$	Yes	Yes	0.556(0.002)	0.703(0.005)	0.529(0.011)	0.749(0.006)	0.649(0.005)	0.801(0.014)
GDEGAN+ESM	$SE(3)$	No	Yes	0.572(0.001)	0.702(0.002)	0.532(0.010)	0.769(0.006)	0.652(0.010)	0.810(0.011)

Table 2: Ablation Study.

Figure 2: Visualization of Protein ‘PDB:1u72(A)’. Left: Model prediction (red) vs true center (green) with coordinates. Right: Predicted residues: True Positive (green), False Positive (red), False Negative (blue).

Figure 3: Attention Patters Visualizations of Protein ‘PDB:3c2f(A)’. Left: On the left we show the attention patterns of GDEGAN, and on the Right: attention patterns of GotenNet(full).

Panel of Figure 4 — Figure 4: Ablation analysis. (a) Model depth analysis showing performance peaks at 4 layers with increased parameters leading to oversmoothing. (b) Temperature evolution across training epochs averaged across multiple runs with mean and standard deviation. (c) Learned temperatures distribution across 8 attention heads (best model).

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Method	Time (s/100 proteins)	Speedup	Type
GDEGAN (Ours)	1.90	19.5×	Reside Level Nodes
GotenNet	4.12	9.0×	Reside Level Nodes
EquiPocket p	37.00	1.0×	Atom Level Nodes
Fpocket p	23.00	1.6×	Geometry Based
Kalasanty p	86.00	0.4×	3D-CNN Based
DeepSurf p	641.00	0.06×	3D-CNN Based

Appendix

Table 3: Inference time comparison across methods.

Figure 5: Training and Validation loss curve. Left: On the left (a) we show training loss convergence, and on the Right: (b) validation loss convergence.

Metric	Value	p-value
Correlation Coefficients
Pearson correlation ( $r$ )	0.3865	$<$ 0.001
Spearman correlation ( $\rho$ )	0.2900	$<$ 0.001
Point-biserial correlation	0.3865	$<$ 0.001
Hypothesis Tests
Mann-Whitney U statistic	1.88 $\times$ 10 9	$<$ 0.001

Appendix

Table 4: Statistical Analysis of Variance-Binding Site Correlation

Figure 6: Variance Distribution: Binding vs Non-binding Residues. Box plot comparing learned neighborhood variance (

\sigma_{i}^{2}

) between binding (red,

n=11{,}107

) and non-binding (blue,

n=194{,}684

) residues on the validation set.

Hyperparameter	Search Space
Learning Rate	{0.003, 0.0003, 0.0005 }
Minimum Learning Rate	1e-6
Batch Size	{8, 16 , 32}
Optimizer	{Adam, AdamW }
Learning rate scheduler	Cosine Annealing Warm Restarts
Warmup Epochs	10

Appendix

Table 5: Hyperparameter selection and reproducibility details.

layers	Param (M)	COACH420		HOLO4K		PDBbind2020
layers	Param (M)	DCC $\uparrow$	DCA $\uparrow$	DCC $\uparrow$	DCA $\uparrow$	DCC $\uparrow$	DCA $\uparrow$
3	1.50	0.528(0.002)	0.673(0.010)	0.534(0.002)	0.731(0.009)	0.643(0.005)	0.777(0.002)
4 (default)	1.90	0.580(0.008)	0.707(0.009)	0.560(0.013)	0.788(0.011)	0.675(0.010)	0.826(0.011)
5	2.30	0.564(0.002)	0.680(0.010)	0.544(0.003)	0.751(0.008)	0.658(0.004)	0.800(0.010)
6	2.70	0.550(0.005)	0.681(0.004)	0.541(0.002)	0.749(0.012)	0.660(0.010)	0.800(0.017)

Appendix

Table 6: Performance of GDEGAN for varying number of message passing layers. All the other parameters were kept same to ensure fair comparison.

Graph Attention Networks

Graph Attention Networks learn how strongly each node should weight information from its neighbors when building graph representations.

See all 75 papers

Multi-Task Genetic Algorithm with Multi-Granularity Encoding for Protein-Nucleotide Binding Site Prediction
Mar 16, 2026Yiming Gao, Liuyi Xu, Pengshan Cui +3Multi-Task Learning Multi-Scale Feature Fusion
College of Information Science and Engineering, Northeastern University, Shenyang, 110819, China · School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China · Key Laboratory of Data Analytics and Optimization for Smart Industry, Ministry of Education, Northeastern University, Shenyang, 110819, China
A Large-Scale Dataset and Benchmark: Do Protein-Ligand Models Learn Binding Sites or Just Binding Likelihood?
May 21, 2026Zhaohan Meng, Zhen Bai, Ke Yuan +4Drug Discovery
School of Computing Science, University of Glasgow · School of Life Science and Technology, Institute of Science Tokyo · School of Cancer Sciences, University of Glasgow +4
Curvature-Informed Potential Energy Surface for Protein-Ligand Binding Affinity Prediction
Jun 12, 2026Peng-Fei Sun, Chuan-Xian Ren, Hong YanProtein-Ligand Binding Affinity Prediction Graph Neural Networks
School of Mathematics, Sun Yat-Sen University, Guangzhou, Guangdong 510275, China · Department of Electrical Engineering, City University of Hong Kong, Hong Kong

Mar 16, 2026J/K move · Enter open · S save

Multi-Task Genetic Algorithm with Multi-Granularity Encoding for Protein-Nucleotide Binding Site Prediction

Yiming Gao, Liuyi Xu, Pengshan Cui, Yining Qian+2

College of Information Science and Engineering, Northeastern University, Shenyang, 110819, China · School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China · Key Laboratory of Data Analytics and Optimization for Smart Industry, Ministry of Education, Northeastern University, Shenyang, 110819, China

Multi-Task Learning Multi-Scale Feature Fusion

Open paper AskPDF arXiv

Accurate identification of protein-nucleotide binding sites is fundamental to deciphering molecular mechanisms and accelerating drug discovery. However, current computational methods often struggle with suboptimal performance due to inadequate feature representation and rigid fusion mechanisms, which hinder the effective exploitation of cross-task information synergy. To bridge this gap, we propose MTGA-MGE, a framework that integrates a Multi-Task Genetic Algorithm with Multi-Granularity Encoding to enhance binding site prediction. Specifically, we develop a Multi-Granularity Encoding (MGE) network that synergizes multi-scale convolutions and self-attention mechanisms to distill discriminative signals from high-dimensional, redundant biological data. To overcome the constraints of static fusion, a genetic algorithm is employed to adaptively evolve task-specific fusion strategies, thereby effectively improving model generalization. Furthermore, to catalyze collaborative learning, we introduce an External-Neighborhood Mechanism (ENM) that leverages biological similarities to facilitate targeted information exchange across tasks. Extensive evaluations on fifteen nucleotide datasets demonstrate that MTGA-MGE not only establishes a new state-of-the-art in data-abundant, high-resource scenarios but also maintains a robust competitive edge in rare, low-resource regimes, presenting a highly adaptive scheme for decoding complex protein-ligand interactions in the post-genomic era.

Abstract

Graph Attention Networks

Explore similar work

Multi-Task Genetic Algorithm with Multi-Granularity Encoding for Protein-Nucleotide Binding Site Prediction

A Large-Scale Dataset and Benchmark: Do Protein-Ligand Models Learn Binding Sites or Just Binding Likelihood?

Curvature-Informed Potential Energy Surface for Protein-Ligand Binding Affinity Prediction