The absence of standardized molecular profiling to differentiate uterine leiomyosarcomas versus leiomyomas represents a current diagnostic challenge. In this study, we aimed to search for a differential molecular signature for these myometrial tumors based on artificial intelligence. For this purpose, differential exome and transcriptome-wide research was performed on histologically confirmed leiomyomas (n = 52) and leiomyosarcomas (n = 44) to elucidate differences between and within these two entities. We identified a significantly higher tumor mutation burden in leiomyosarcomas vs. leiomyomas in terms of somatic single-nucleotide variants (171,863 vs. 81,152), indels (9491 vs. 4098), and copy number variants (8390 vs. 5376). Further, we discovered alterations in specific copy number variant regions that affect the expression of some tumor suppressor genes. A transcriptomic analysis revealed 489 differentially expressed genes between these two conditions, as well as structural rearrangements targeting ATRX and RAD51B. These results allowed us to develop a machine learning approach based on 19 differentially expressed genes that differentiate both tumor types with high sensitivity and specificity. Our findings provide a novel molecular signature for the diagnosis of leiomyoma and leiomyosarcoma, which could be helpful to complement the current morphological and immunohistochemical diagnosis and may lay the foundation for the future evaluation of malignancy risk.
Keywords: classification model; diagnostic/prognostic biomarkers; differential gene expression; exome/transcriptome; integrative analysis; leiomyoma; leiomyosarcoma; machine learning; mutational pattern.