A statistical framework for genomic data fusionGert R. G. Lanckriet, Tijl De Bie, Nello Cristianini, Michael I. Jordan and William Stafford Noble.
Data and kernel matricesIn the following table, the data and kernel matrices are provided as tab-delimited text files, with string IDs in the first row and column. Warning: Some of the kernel matrices are quite large and may cause your browser to hang if you try to load them directly. You can save them to your local disk by right clicking on the link.
The links marked "PNG" are heat map representations of the corresponding matrices, generated using matrix2png. For the data matrices, the color range is selected to span the middle 98% of the values in the matrix.
In order to evaluate the kernel matrices for the prediction problems further on, the relevant submatrices (corresponding to the genes with known classification) need to be selected, and then either normalized (N) or centered and then normalized (CN), as indicated in the last column of the table. When predicting unannotated genes or proteins, the full kernel matrices are needed and should accordingly be centered or not and then normalized.
Matlab code to center a kernel matrix can be found here and Matlab code for normalization here.
An example membrane protein. pSR II is a photoreceptor in the plasma membrane of archaeon Natronobacteriumpharaonis and functions as a sensor for phototactic avoidance.