Figure 2.

eScience-Bayes applied to microarray gene expression data. Illustration of using the eScience-Bayes approach to model time to development of distant metastases in breast cancer patients from microarray gene expression data. (Color-coding is as in Fig. 1; the green blocks in Fig. 1 are not included since we did not use any locally produced data or information.) (A) We downloaded five breast cancer gene expression datasets and their associated clinical data using the Gene Expression Omnibus (GEO) Web service. For trustworthy prior information about genes with altered transcriptional regulation in breast cancer patients, we used three Web services in conjunction: NetPath, DictService and Entrez Utilities. This gave us a list of integers that represented the number of times a cancer-pathway regulated gene on the HG-U133A array was mentioned in PubMed articles together with breast cancer. (B) We used the prior information derived in (A) to restrict the dimensionality of Models I and II by Bayesian variable selection. We did this by deriving prior distributions based on the assumption that the probability of a gene being related to breast cancer was reflected in the number of times a gene reported in NetPath was mentioned in combination with breast cancer in PubMed articles. (C) The models were fit by performing calculations a HPC resource.