Digital Total - Computing & Data Science an der Universität Hamburg und in der Wissenschaftsmetropole Hamburg

Identification of differentially expressed biclusters for unsupervised patient stratification

95

Nicht eingeplant

20m

Von-Melle-Park 4

Poster

Beschreibung

Unexplored molecular heterogeneity of human diseases causes treatment inefficacy and hinders the investigation of causative disease mechanisms. Since the number and frequencies of disease subtypes are usually unknown, unsupervised methods are applied to omics data to identify patients subgroups with similar molecular profiles. Here, we present UnPaSt, a novel biclustering algorithm for unsupervised patient stratification and demonstrate its superior performance compared to traditionally used clustering, factorization, and biclustering methods in benchmarks with simulated and real data. Moreover, besides accurate identification of well-known PAM50 subtypes of breast cancer, UnPaSt detected the rare neuroendocrine subtype, which was overlooked in previous analyses due to its low frequency.

Keywords

patient stratification
clustering
biclustering
disease heterogeneity
omics