作者: Ross D. Markello , Golia Shafiei , Christina Tremblay , Ronald B. Postuma , Alain Dagher
DOI: 10.1101/2020.03.05.979526
关键词: Patient characteristics 、 Computational biology 、 Parkinson's disease 、 Disease 、 Unsupervised learning 、 Computer science 、 Cognition 、 Neuroimaging 、 Similarity (psychology)
摘要: Individuals with Parkinson9s disease present a complex clinical phenotype, encompassing sleep, motor, cognitive, and affective disturbances. However, characterizations of PD are typically made for the "average" patient, ignoring patient heterogeneity obscuring important individual differences. Modern large-scale data sharing efforts provide unique opportunity to precisely investigate characteristics, but there exists no analytic framework comprehensively integrating modalities. Here we apply an unsupervised learning method---similarity network fusion---to objectively integrate MRI morphometry, dopamine active transporter binding, protein assays, measurements from n = 186 individuals de novo Progression Markers Initiative. We show that multimodal fusion captures inter-dependencies among modalities would otherwise be overlooked by field standard techniques like concatenation. then examine how subgroups derived fused map onto phenotypes, neuroimaging is critical this delineation. Finally, identify compact set phenotypic axes span population, demonstrating continuous, low-dimensional projection patients presents more parsimonious representation in sample compared discrete biotypes. Altogether, these findings showcase potential similarity combining heterogeneous populations.