Exploring local patterns between gene expression profiles and chemical structures (fingerprints) of compounds

Abstract

The availability of high throughput technologies such as microarrays and next generation sequencing have made it possible to cheaply collect large amounts of drug-gene expression data sets. Combining compounds and their characteristics with gene expression data is called connectivity mapping and holds promise for in-depth analysis and understanding of biological processes, discovery of new drug targets and new drugs and prediction of toxic potential of unknown compounds. These goals can be achieved using the connectivity map data base and using appropriate methods.

For studying relationships between gene expression profiles of human cells following the introduction of chemical compounds and the fingerprints of the compounds, the recently developed Multiple factor analysis (MFA) which seeks patterns in data consisting of quantitative as well as qualitative variables can be applied. The results of the MFA can often be made more robust by applying hierarchical clustering analysis. In addition, ignoring the gene expression profiles and working only with fingerprints of compounds, it was determined whether groups of compounds are associated with groups of fingerprints using five different methods: Multiple correspondence analysis (MCA), Binary inclusion-maximal biclustering (Bimax) algorithm, Factor analysis for Bi-cluster acquisition (FABIA), Iterative Binary biclustering of gene sets and Factor analysis for binary data. These biclustering approaches simultaneously cluster rows and columns of the data matrix.

The perfo