Recent progress in epitope prediction has shown promising results in the development of vaccines and therapeutics against various diseases. However, the overall accuracy and success rate need to be improved greatly to gain practical application significance, especially conformational epitope prediction. In this review, we examined the general features of antibody–antigen recognition, highlighting the conformation selection mechanism in flexible antibody–antigen binding. We recently highlighted the success and warning signs of antibody epitope predictions, including linear and conformation epitope predictions. While deep learning-based models gradually outperform traditional feature-based machine learning, sequence and structure features still provide insight into antibody–antigen recognition problems.

T-cell receptors (TCRs) recognize peptide–MHC (pMHC) complexes through intricate structural interactions, which is a core component of adaptive immunity. However, the diverse and cross-reactive nature of TCRs poses great challenges for accurate prediction of TCR-epitope interactions, hampering the advancement and broad application of TCR-related therapies. Here, we present SageTCR, a bi-level graph neural network (GNN) framework that leverages structural data to predict TCR-pMHC binding possibilities. Harnessing the pretrained language models, SageTCR encodes detailed structural arrangement at both residue-level and atomic-level and effectively integrates the bimodal representations via attention mechanisms. To tackle the deficiency of experimental structures, we explore comprehensive data augmentation strategies to enrich the training and increase the generalizability while concurrently preserving the characteristic TCR-pMHC diagonal binding mode. SageTCR demonstrates superior performance compared to six methods with different deep learning architectures. Furthermore, SageTCR offers the interpretability by identifying and focusing on the conformational features of pivotal contact residues on the interface, which can provide valuable insights for TCR engineering and immunotherapy design.

Focal adhesion kinase (FAK) is a kind of tyrosine kinase that modulates integrin and growth factor signaling pathways. FAK is a promising therapeutic target, involving in cancer cell migration, proliferation and survival. In order to design highly active FAK inhibitors, a three-dimensional quantitative structure–activity relationship (3D-QSAR), molecular docking and molecular dynamics simulation were combined to study the interaction between pyrimidine derivatives and FAK. Both the comparative molecular field analysis (CoMFA) model (q² = 0.664, r² = 0.973) and comparative molecular similarity index analysis (CoMSIA) model (q² = 0.755, r² = 0.999) indicated that the constructed QSAR models are stable and reliable. Based on the model, several new compounds were designed and their inhibitory activities were validated by molecular docking and molecular dynamics simulation. The results show that the complex of FAK with new designed compounds is stable during the simulation, which supports the fact that the new compounds are potential FAK inhibitors.

Serum and glucocorticoid-regulated protein kinase 1 (SGK1) that can promote the growth of tumor cells is highly expressed in many tumors. Sarcoma gene (Src), a class of oncogenes, plays an important role in the pathogenesis of cancer and is an important kinase in tumor cell expression pathways. Research has shown that SGK1, as a key regulator, is involved in Src-induced tumor formation, and tumor progression can be inhibited by inhibiting SGK1/Src in combination. In this study, SGK1/Src dual inhibitors were designed by computer-aided drug design. Seven million molecules were screened via the pharmacophore model and molecular docking. Finally, 6 compounds were selected. The results of molecular dynamics simulation of compound 1 showed that it has good binding stability to both receptors. According to the binding site of compound 1 with two receptors, corresponding modifications were made, and 6 novel designed compounds were obtained. The docking energies of the novel designed compounds were lower than that of compound 1.

Predicting Antibody–Antigen (Ab–Ag) docking and structure-based design represent significant long-term and therapeutically important challenges in computational biology. We present SAGERank, a general, configurable deep learning framework for antibody design using Graph Sample and Aggregate Networks. SAGERank successfully predicted the majority of epitopes in a cancer target dataset. In nanobody–antigen structure prediction, SAGERank, coupled with a protein dynamics structure prediction algorithm, slightly outperforms Alphafold3. Most importantly, our study demonstrates the real potential of inductive deep learning to overcome the small dataset problem in molecular science. The SAGERank models trained for antibody–antigen docking can be used to examine general protein–protein interaction tasks, such as T Cell Receptor-peptide-Major Histocompatibility Complex (TCR-pMHC) recognition, classification of biological versus crystal interfaces, and prediction of ternary complexes of molecular glues. In the cases of ranking docking decoys and identifying biological interfaces, SAGERank is competitive with or outperforms state-of-the-art methods.