Analysis of Genomic and Proteomic Data Using Advanced Literature Mining

High-throughput technologies, such as proteomic screening and DNA micro-arrays, produce vast
amounts of data requiring comprehensive analytical methods to decipher the biologically relevant
results. One approach would be to manually search the biomedical literature; however, this would be
an arduous task. We developed an automated literature-mining tool, termed MedGene, which
comprehensively summarizes and estimates the relative strengths of all human gene−disease
relationships in Medline. Using MedGene, we analyzed a novel micro-array expression dataset
comparing breast cancer and normal breast tissue in the context of existing knowledge. We found no
correlation between the strength of the literature association and the magnitude of the difference in
expression level when considering changes as high as 5-fold; however, a significant correlation was
observed (r = 0.41; p = 0.05) among genes showing an expression difference of 10-fold or more.
Interestingly, this only held true for estrogen receptor (ER) positive tumors, not ER negative. MedGene
identified a set of relatively understudied, yet highly expressed genes in ER negative tumors worthy of
further examination.
Keywords: bioinformatics • micro-array • text mining • gene-disease association • breast cancer