Multiscale Element-Doped Nanowire Array-Coupled Machine Learning Reveals Metabolic Fingerprints of Nonreversible Liver Diseases.

Timely detection of nonreversible liver diseases contributes greatly to reasonable therapy and quality of life. Given the current situation, minimally invasive high-specificity molecular diagnosis based on body fluid can be a good choice. Herein, a mesoporous superstructure is designed using silicon atom-doped nanowire arrays to uniformly load Pt nanoparticles on the surface to produce a desirable ionization effect. We apply the multiscale element-doped nanowire arrays to efficiently assist extraction of high-quality metabolic fingerprints from only 35 nL of serum within seconds. Using different machine learning algorithms, we establish specific biomarker panels to distinguish different liver diseases from the healthy control, with more than 90% accuracy, sensitivity, and specificity. Moreover, from established biomarker panels, we further determine key metabolites of significant difference ( p < 0.01) via group comparison to realize the discrimination of different liver diseases with 100% sensitivity. Our work confirms the design protocol of an advanced diagnosis tool and lays a robust foundation for metabolic molecular diagnosis in large-scale clinical application.