Genomic foundation models extend clinicopathologic and transcriptomic prognostication in soft tissue sarcoma

The study tested whether the Evo2 genomic foundation model improves prognostication of soft tissue sarcomas compared to existing clinicopathologic and transcriptomic models. The research included 102 patients with three types of sarcomas (dedifferentiated liposarcoma, high-grade leiomyosarcoma, and undifferentiated pleomorphic sarcoma) and an external validation cohort of 129 patients from the TCGA database. Genomic alterations were converted into sequence inputs and processed using the Evo2 model. Results showed significant improvements: in the study cohort, the concordance index for overall survival increased from 0.620 to 0.770 when combining Evo2 with the clinicopathologic model. In the TCGA cohort, the concordance index for overall survival increased from 0.595 to 0.716. The Evo2 model provided independent prognostic information and outperformed individual existing models. The authors concluded that genomic foundation models represent a complementary layer for sarcoma risk stratification, but confirmation of clinical utility is required.