Table 7.

Summary of Contributions in Studies on Attention and Transformer for Protein-Protein Interactions. Note that each study employed varied datasets, cross-validation methods, and simulation settings for evaluation, making direct comparisons potentially inconclusive. The highest reported accuracy is presented when models were assessed using multiple datasets.

Author	Metrics and Results	Contributions
Asim et al. [91]	ACC: 0.926 F1: 0.9195 MCC: 0.855	Proposed ADH-PPI, an attention-based hybrid model with superior accuracy for PPI prediction.
Baek et al. [92]	ACC: 0.868 MCC: 0.768 F1: 0.893 AUC: 0.982	Utilized a three-track neural network integrating information at various dimensions for protein structure and interaction prediction.
Li et al. [93]	F1: 0.925	Offered a PPI relationship extraction method through multigranularity semantic fusion, achieving high F1-scores.
Li et al. [94]	ACC: 0.9519 MCC: 0.9045 AUC: 0.9860	Introduced SDNN-PPI, a self-attention-based PPI prediction method, achieving up to 100% accuracy on independent datasets.
Nambiar et al. [95]	ACC: 0.98 AUC: 0.991	Developed a Transformer neural network that excelled in protein interaction prediction and family classification.
Tang et al. [96]	ACC: 0.631 F1: 0.393	Proposed HANPPIS, an effective hierarchical attention network structure for predicting PPI sites.
Warikoo et al. [97]	F1: 0.86	Introduced LBERT, a lexically aware transformer-based model that outperformed state-of-the-art models in PPI tasks.
Wu et al. [98]	AUPRC: 0.8989	Presented CFAGO, an efficient protein function prediction model integrating PPI networks and protein biological attributes.
Zhang and Xu [99]	ACC: 0.856	Introduced a kernel ensemble attention method for graph learning applied to PPIs, showing competitive performance.
Zhu et al. [100]	ACC: 0.934 F1: 0.932 AUC: 0.935	Introduced the SGAD model, improving the performance of Protein Interaction Network Reconstruction.