Open Collections

A systems biology approach for identifying markers of chemotherapy response

University of British Columbia

High-throughput gene expression data has been widely used to identify biomarkers for the classification of clinical outcome in cancer studies. In breast cancer, conventional methods have successfully identified molecular markers predictive of disease progression; however, predicting response to chemotherapy has proved more challenging and warrants the development of novel approaches. Recently developed systems biology methods that integrate transcriptomic and proteomic data have shown promising results in various classification problems; therefore, we investigated the use of this approach in predicting response to chemotherapy. We developed a novel method, called OptDis, which integrates gene expression data with protein-protein interaction networks to efficiently identify subnetwork markers with optimal discrimination between different clinical outcome groups. Application of our method to a public dataset demonstrated three key advantages of using OptDis over previous methods for predicting drug response in breast cancer patients treated with combination chemotherapy. First, subnetwork markers derived from our method provides better classification performance compared with subnetwork and gene marker from existing methods. Second, OptDis subnetwork markers are more reproducible across independent cohorts compared to gene markers and may consequently be more robust against noise and variations in expression data. Third, OptDis subnetwork markers provide insights into mechanisms underlying tumour response to chemotherapy that are missed by conventional methods. Additional analyses using OptDis showed that the use of prior knowledge from PPI interactions improves marker discovery and subsequent classification performance. To our knowledge, this is the first study to demonstrate the advantages of applying an integrative network-based approach to the prediction of individual’s response to cancer treatment. Markers identified using our method not only improve the classification of outcome, but it also provide novel understandings into the mechanism of drug action. With sufficient validation, this strategy may identify promising clinical markers that can facilitate the effective individualised treatment of cancer patients.

Text

2012-04-20

Attribution 3.0 Unported

http://hdl.handle.net/2429/42134

Bioinformatics

University of British Columbia

UBCV

http://creativecommons.org/licenses/by/3.0/