Harnessing the LINCS L1000 data for drug discovery by computational analyses.

摘要

More than 1.4 million gene expression profiles have been produced for the Library of Integrated Network-based Cellular Signatures (LINCS) project using the L1000 technology. Processing and analyzing such big dataset for new biomedical insights poses unprecedented challenges and opportunities. For my thesis work I applied a novel differential expression method called the characteristic direction (CD) to compute differential expression signatures from the LINCS L1000 data. The calculated signatures show better performance than existing methods to compute signatures. Through several unique benchmarking schemes, I demonstrated to have a better way to maximally extract biological knowledge from the LINCS L1000 data. With this improved method, I developed the LINCS L1000 characteristic direction signature search (L1000CDS 2) engine. L1000CDS2 is an online web application that enables users to search for LINCS L1000 small molecule signatures that mimic or reverse an input signature. In collaboration with the United States Army Medical Research Institute of Infectious Diseases (USAMRIID), using the tool I developed, we identified Kenpaullone, a GSK3B/CDK2 kinase inhibitor, as a potent inhibitor of EBOV infection in human cell lines. In another collaborative project with investigators at Harvard Medical School, I analyzed CD signatures of six breast cancer cell lines treated with anti-cancer compounds. I found that there are both common and cell type specific transcriptional responses that are generally correlated with phenotypical cell imaging data that are used to assess growth rates. Together, we also discovered that orthogonality of two CD signatures is predictive of drug synergy, and can be used as a way to find effective drug combinations in cancer. To facilitate the use of the LINCS L1000 data by other researchers, I also developed several online interactive web applications. These tools can help users to navigate through this new big dataset, visualize selected gene expression profiles on canvases or 3D plots, and perform enrichment analyses to identify connections between gene expression profiles and existing biological knowledge. So far over 5,000 users utilized the web based tools that I developed. Overall, this thesis work represents the first systematic study and application of the LINCS L1000 dataset for novel discoveries. It highlights the opportunity big data present to advance translational biomedical research.