Novel implementation of conditional co-regulation by graph theory to derive co-expressed genes from microarray data

Abstract

Abstract Background Most existing transcriptional databases like Comprehensive Systems-Biology Database (CSB.DB) and Arabidopsis Microarray Database and Analysis Toolbox (GENEVESTIGATOR) help to seek a shared biological role (similar pathways and biosynthetic cycles) based on correlation. These utilize conventional methods like Pearson correlation and Spearman rank correlation to calculate correlation among genes. However, not all are genes expressed in all the conditions and this leads to their exclusion in these transcriptional databases that consist of experiments performed in varied conditions. This leads to incomplete studies of co-regulation among groups of genes that might be linked to the same or related biosynthetic pathway. Results We have implemented an alternate method based on graph theory that takes into consideration the biological assumption – conditional co-regulation is needed to mine a large transcriptional data bank and properties of microarray data. The algorithm calculates relationships among genes by converting discretized signals from the time series microarray data (AtGenExpress) to output strings. A 'score' is generated by using a similarity index against all the other genes by matching stored strings for any gene queried against our database. Taking carbohydrate metabolism as a test case, we observed that those genes known to be involved in similar functions and pathways generate a high 'score' with the queried gene. We were also able to recognize most of the randomly selected correlated pairs from Pearson correlation in CSB.DB and generate a higher number of relationships that might be biologically important. One advantage of our method over previously described approaches is that it includes all genes regardless of its expression values thereby highlighting important relationships absent in other contemporary databases. Conclusion Based on promising results, we understand that incorporating conditional co-regulation to study large expression data helps us identify novel relationships among genes. The other advantage of our approach is that mining expression data from various experiments, the genes that do not express in all the conditions or have low expression values are not excluded, thereby giving a better overall picture. This results in addressing known limitations of clustering methods in which genes that are expressed in only a subset of conditions are omitted. Based on further scope to extract information, ASIDB implementing above described approach has been initiated as a model database. ASIDB is available at http://www.asidb.com.