Diagnosing Disease

For some diseases, expression profiling is a method of choice for identifying marker genes able to diagnose the severity of disease and to predict future disease outcomes. Marker genes are selected by scoring how well their expression levels can discriminate between different classes of disease. This method has achieved >90% accuracy for leukemias such as acute myeloid leukemia and acute lymphoblastic leukemia. For other cancers, however, mRNA-based classification has yet to achieve high accuracy. One reason why expression-based diagnostics may fail is that changes in expression of the relatively few genes causing disease may be subtle compared to those of the downstream effectors which may vary considerably from patient to patient.

Many groups have hypothesized that a more effective means of marker identification may be to combine gene expression measurements over groups of genes that fall within common pathways. In past years, the Ideker lab has developed several approaches for integrating expression profiles with pathways curated from the literature or extracted from protein interaction networks [Chuang et al. 2007]. Large protein-protein interaction networks have only recently become available for human, enabling new opportunities for elucidating pathways involved in major diseases and pathologies. Our methodology is to overlay a patient's expression profile onto a network map of the cell to identify subnetworks and pathways that are predictive of disease. This approach has shown success in diagnosis of metastatic breast cancer, and we are now working together with Dr. Thomas Kipps at the UCSD Moores Cancer Center to diagnose aggressive versus indolent cases of Chronic Lymphocytic Leukemia (CLL).

Diagnosing disease in the lab is funded by Aglient, Pfizer, and IIS-0803937.

Interpreting personalized genetic information

Although genome-wide association studies (GWAS) are rapidly increasing in number, numerous challenges persist in identifying and explaining the associations between loci and quantitative phenotypes. This project is developing tools to integrate gene association data with protein network information to identify the pathways underlying a patient’s genotype.  These methods will elevate the study of gene association to a new study of “pathway association.”  The project is a joint work with Richard Karp in the EECS Department at UC Berkeley.

Our proposed solution is to explain the associations captured by GWAS in terms of known gene and protein interactions.  New technologies have provided a wealth of interaction data ranging from the proteome (protein-protein interaction networks) to the transcriptome (protein-DNA interactions) to the metabolome (metabolic pathways). We will develop computational tools that query these independent networks to identify pathways and sub-networks of interactions underlying the observed set of genome-wide associations.  This framework is intended to improve the power of current GWAS, by identifying genes in loci with borderline significance that nonetheless have close network proximity to significant genes.  Furthermore, it will provide a list of putative physical pathways incorporating the causal genes necessary to affect the phenotype. Some of our early efforts along these lines are available in Suthram et al. 2008.

Intrepretation of personalized genetic information in the lab is funded by a grant from Microsoft.