体细胞突变的选择景观在乳腺癌中编码组织病理学信息

Cancer effect size quantifies the strength of evolutionary selection acting on somatic mutations within tumors, providing insights into which genetic alterations are being actively selected during tumor evolution. While histopathology has long been the gold standard for cancer diagnosis and classification, the relationship between evolutionary selection pressures on specific mutations and observable tumor morphology remains unexplored. We investigated whether combinations of somatic mutations, weighted by their cancer effect sizes, could serve as genomic proxies for pathological features in breast cancer. Using deep learning-based feature extraction from pre-trained histopathology models, we extracted high-dimensional morphological representations from H&E-stained whole slide images and performed cross-modal analysis to identify which mutation-selection profiles correspond to distinct histological patterns. By identifying mutation combinations whose evolutionary selection signatures associate with specific pathological features, we tested whether these genomic profiles could contribute to classifying breast cancer molecular subtypes. Our findings reveal that evolutionary selection patterns acting on specific mutation combinations show measurable correspondence with pathological features and provide contributory predictive value for subtype classification, though they do not fully recapitulate the discriminative power of histopathological assessment. This suggests that the selective landscape of somatic mutations partially encodes morphological information, with cancer effect sizes capturing aspects of the evolutionary processes that shape tumor architecture. These results provide evidence that mutation-selection profiles can complement traditional pathology and advance our understanding of how evolutionary forces influence both the genetic and morphological characteristics of breast cancer. Moreover, because cancer effect sizes capture the directionality and strength of selective pressures, their association with histopathological assessment models may improve prediction of tumor evolutionary trajectories without -omics data, offering insights into likely progression patterns with which to inform treatment strategy.