Recent advancements in tumor bioinformatics have been spearheaded by researchers from the Computer Network Information Center of the Chinese Academy of Sciences (CAS). They have leveraged the "Eastern" supercomputing system to develop a novel method called SuRe-Transformer, which uses Transformer architecture to predict homologous recombination deficiency (HRD) in breast cancer pathology images. This innovative approach addresses three primary challenges in applying Transformer models to pathology image analysis. First, it introduces a radial decay sparse self-attention mechanism to enhance the diversity of selected patches, ensuring that critical regions are adequately represented. Second, the method employs a cluster size-weighted sampling strategy to improve the ability to capture representative features from key areas. Lastly, it incorporates unsupervised contrastive learning to refine the quality of feature embeddings derived from the pathology images. Experimental results demonstrate that SuRe-Transformer outperforms existing mainstream techniques in HRD prediction across multiple independent test sets. Its robustness and versatility are further underscored by its superior performance in predicting mutations in several important genes, including TP53, PIK3CA, and MAP3K1, with an area under the curve (AUC) exceeding 0.844 in each case. The research has been accepted for publication in npj Precision Oncology, a leading journal in the field. This work was supported by the National Natural Science Foundation of China and the Chinese Academy of Sciences' Strategic Priority Research Program (Class B). The development of SuRe-Transformer marks a significant step forward in the application of deep learning techniques to tumor pathology, potentially enhancing the accuracy and efficiency of cancer diagnostics and treatments. By addressing key limitations of previous methods, this model paves the way for more reliable genetic mutation predictions, which are crucial for personalized medicine and targeted therapies.