nov
2018
Mamuten , Medicinaregatan 18A Plan 4
Genomic data calls for integrative analysis vertically (groups of patients) and horizontally (data types or modalities). The complexity of such genomic data is ever increasing.
The patient cohorts can correspond to different disease states or studies, the modalities to different kinds of ‘omics data. We can have access to clinical information,
pharmacological data and structured data base information. There is also a possibility that only a subset of data type information is available for each cohort.
We propose MV-PCA as a new method for multiview data integration. The methods builds on a regularized factor analysis with sparsity constraints on both loadings and
scores, and group constraints to allow for partial and interpretable data integration in both the horizontal and vertical direction. The method thus identifies which patient subgroups
have shared structure for which set of variables, across both cohorts and data types simultaneously.
We demonstrate MV-PCA on simulated data and present some applications to genomic data.
Speaker: Rebecka Jörnsten, Professor of Biostatistics and Applied Statistics, Mathematical Sciences, Chalmers University of Technology/University of Gothenburg