25 May 2022 in Computer Methods and Programs in Biomedicine
4 June 2025 in Nature Communications
Precise identification of T cell antigens in silico is crucial for the development of cancer mRNA vaccines. However, current computational methods only utilize sequence-level rather than atomic level features to identify T cell antigens, which results in poor representation of those that activate immune responses. Here we propose deepAntigen, a graph convolutional network-based framework, to identify T cell antigens at the atomic level. deepAntigen achieves excellent performance both in the prediction of antigen-human leukocyte antigen (HLA) binding and antigen-T cell receptor (TCR) interactions, which can provide comprehensive guidance for identification of T cell antigens. The tumor neoantigens predicted by deepAntigen in lung, breast and pancreatic cancer patients are experimentally validated through ELISPOT assays, which detect successful activation of CD8+ T cells to release IFN-γ. Overall, deepAntigen can accurately identify T cell antigens at the atomic level, which could accelerate the development of personalized neoantigen targeted immunotherapies for cancer patients.
3 May 2023 in Expert Systems with Applications
13 August 2024 in Expert Systems with Applications
1 July 2026 in IEEE Journal of Biomedical and Health Informatics
Breast cancer is the leading cause of death among women worldwide. Ultrasound imaging is an important means for the early detection of breast cancer, improving the survival rate. Due to the shortage of experienced sonographers, computer-aided systems for breast cancer recognition become particularly important. Some recent studies analyze tumor types in lesion regions but rely on predefined ROIs. Some other studies recognize cancer in the whole ultrasound image, but always suffer from the extremely variable proportion, location and quantity of the tumor lesions. In this paper, we propose a regional prior distillation (RPD) framework for breast cancer diagnosis in ultrasound images. To enhance the analysis of the tumor region, we propose an Image-Cross Attention (ICA) to fuse the predefined ROI prior information with ultrasound images and train a prior-fused model. To remove the constraint of predefined ROIs, we propose a Distribution Distillation Learning (DDL) to distill the prior-fused sample distribution from the prior-fused model into a diagnostic model, which analyzes the disease from only ultrasound images, based on the knowledge distillation paradigm of the teacher-student framework. Comprehensive experiments are conducted on multi-institutional datasets to validate the proposed RPD framework. The results demonstrate the following points. The ICA fuses regional prior information adequately, leading to a high-performance prior-fused model. The DDL distills the prior information effectively, enhancing the diagnostic model to focus on the tumor lesions. The performance of the diagnostic model surpasses that of current SOTA methods by 1.66% in accuracy and 0.64% in AUC. In addition, the diagnostic model is robust to slight perturbations and achieves good generalization performance.
2 July 2025 in Briefings in Bioinformatics
Spatial transcriptomics (ST) encompasses rich multi-modal information related to cell state and organization. Precisely identifying spatial domains with consistent gene expression patterns and histological features is a critical task in ST analysis, which requires comprehensive integration of multi-modal information. Here, we propose TriCLFF, a contrastive learning-based multi-modal feature fusion framework, to effectively integrate spatial associations, gene expression levels, and histological features in a unified manner. Leveraging an advanced feature fusion mechanism, our proposed TriCLFF framework outperforms existing state-of-the-art methods in terms of accuracy and robustness across four datasets (mouse brain anterior, mouse olfactory bulb, human dorsolateral prefrontal cortex, and human breast cancer) from different platforms (10x Visium and Stereo-seq) for spatial domain identification. TriCLFF also facilitates the identification of finer-grained structures in breast cancer tissues and detects previously unknown gene expression patterns in the human dorsolateral prefrontal cortex, providing novel insights for understanding tissue functions. Overall, TriCLFF establishes an effective paradigm for integrating spatial multi-modal data, demonstrating its potential for advancing ST research. The source code of TriCLFF is available online at https://github.com/HBZZ168/TriCLFF.