Machine Learning-Derived Sarcopenia Signature Identifies High-Risk Molecular State in T2D Skeletal Muscle.

Abstract

Sarcopenia is common in type 2 diabetes (T2D), but whether sarcopenia-associated transcriptomic features can be detected in diabetic skeletal muscle remains unclear. A machine-learning signature combining 26 gene-expression features and one ssGSEA-derived aggregate feature was developed in GSE111016 and GSE226151 and independently evaluated in GSE111010. The frozen signature was projected onto T2D bulk transcriptomic cohorts and a nine-donor single-nucleus RNA-sequencing dataset. Its association with lower-extremity physical performance was further evaluated in GSE144304. The model achieved AUCs of 0.906 in the training cohort and 0.762 in the validation cohort. In T2D skeletal muscle, the high-risk molecular state was associated with inflammatory, extracellular-matrix, adhesion, and stress-response programs, with SESN3 and VCAM1 identified as candidate genes associated with shared pathological processes. In GSE144304, a higher frozen signature score was associated with lower SPPB after adjustment for age, sex, and BMI. Single-nucleus analysis provided exploratory cellular context for potential variation in cell-type and myonuclear distributions. The frozen molecular signature can identify a high-risk molecular state in T2D skeletal muscle.