Abstract
BACKGROUND: High-throughput experiments, such as with DNA microarrays, typically result in hundreds of genes potentially relevant to the process under study, rendering the interpretation of these experiments problematic. Here, we propose and evaluate an approach to find functional associations between large numbers of genes and other biomedical concepts from free-text literature. For each gene, a profile of related concepts is constructed that summarizes the context in which the gene is mentioned in literature. We assign a weight to each concept in the profile based on a likelihood ratio measure. Gene concept profiles can then be clustered to find related genes and other concepts. RESULTS: The experimental validation was done in two steps. We first applied our method on a controlled test set. After this proved to be successful the datasets from two DNA microarray experiments were analyzed in the same way and the results were evaluated by domain experts. The first dataset was a gene-expression profile that characterizes the cancer cells of a group of acute myeloid leukemia patients. For this group of patients the biological background of the cancer cells is largely unknown. Using our methodology we found an association of these cells to monocytes, which agreed with other experimental evidence. The second data set consisted of differentially expressed genes following androgen receptor stimulation in a prostate cancer cell line. Based on the analysis we put forward a hypothesis about the biological processes induced in these studied cells: secretory lysosomes are involved in the production of prostatic fluid and their development and/or secretion are androgen-regulated processes. CONCLUSION: Our method can be used to analyze DNA microarray datasets based on information explicitly and implicitly available in the literature. We provide a publicly available tool, dubbed Anni, for this purpose.