Skip to main content
Fig. 2 | Genome Medicine

Fig. 2

From: Extracting a low-dimensional description of multiple gene expression datasets reveals a potential driver for tumor-associated stroma in ovarian cancer

Fig. 2

Overview of the evaluation and application of INSPIRE procedure. INSPIRE takes as input K ≥2 datasets, and the method is an iterative procedure that determines the assignment of the genes to modules, the features each corresponding to a module, and the dependencies among the features which are conserved across the datasets. a Evaluation of INSPIRE using simulated data. Two simulated datasets in four settings corresponding to different amount of gene overlap are provided as input to the INSPIRE learning algorithm, and the learned modules and network are evaluated in terms of three different metrics. b Evaluation of INSPIRE using two ovarian cancer expression datasets. Two expression datasets from different platforms are provided as input to the learning algorithm and the learned modules and network are evaluated in terms of three different metrics. c Application of INSPIRE on nine real-world ovarian cancer expression datasets. As an application of INSPIRE, we first check the association of the learned INSPIRE features with six histological and clinical phenotypes, which is followed by subtyping the patients into groups based on the learned INSPIRE features. Observing that INSPIRE features have high association with the histological and clinical phenotypes in cancer and the subtypes learned based on the features can predict copy number variation (CNV) abnormalities well leads us to do a deeper analysis of two modules (modules 5 and 6), which are good predictors of many phenotypes and good differentiators of learned ovarian cancer subtypes

Back to article page