FacPad is the inference of Pathways responsive to drug treatment via Bayesian sparse Factor modeling.
It requires two matrices as input datasets. The first matrix Y is the gene expression ratios before and after drug treatment. It has dimension G〜J, where G is the number of probesets measured by the specific microarray platform and J is the number of different treatments (usually, a treatment is combination of certain drug, concentration, and treatment time). The second matrix L is the binary pathway structure matrix of dimension G〜K, where K is the number of pathways associated with the probesets, with Lg,k=1 representing that the g-th probeset is mapped to the k-th pathway and Lg,k=0 otherwise. FacPad models each pathway as a latent factor which is the weighted combination of its associated probesets, and decomposes the matrix Y into loading matrix W (G〜K) and factor activity matrix X (K〜J):
Figure. Illustration of the bayesian sparse factor model used for the analysis of treatment response data.
The sparse structure of loading matrix W is determined by prior binary matrix L.
Dataset
MIcroarray expression data before and after drug treatment was downloaded from the Connectivity Map database:http://www.broadinstitute.org/cmap/cel_file_chunks.jsp
Pathways associated with the probesets were derived using the functional annotation function of the Database for Annotation, Visualization and Integrated Discovery (DAVID):
http://david.abcc.ncifcrf.gov/summary.jsp
Software
The bayesian sparse factor model is implemented as the R package "FacPad" on CRAN:
http://cran.open-source-solution.org/web/packages/FacPad/