Mission Statement

Genetic variation is critical to our susceptibility to diseases and response to medications. Yet the functional consequences of most genetic variants are unknown. We are working to predict these consequences using computation, by integrating information from molecular modeling and sequence analysis with clinical patient data and in vitro functional studies, through collaborations with physicians, genetic counselors, and experimental biologists. We are particularly interested in inherited cancer susceptibilities and gain of function mutations in tumor genomes.

Computational analysis of protein evolution and structure to aid medical decision making in high-risk breast, ovarian, prostate, and colorectal cancer families

In the new "post-genome" era of personalized medicine, many genes critical to disease susceptibilities and drug sensitivies will be identified and increased numbers of people will undergo genetic testing. Already, thousands of people from high risk families are being tested for known pathogenic mutations in genes associated with cancer.

We are developing a Bayesian approach to classification of germline variants in cancer risk genes that can be used in a clinical setting. Currently, a prior probability of variant cancer risk is updated by two independent likelihood ratios to produce a posterior probability of cancer risk. A family history likelihood is combined with a protein based likelihood ratio that estimates the probability of a variant causing a loss of protein function. Loss of function is predicted based on protein tertiary structure, evolutionary conservation and amino acid properties. The method gives clinicians a likelihood ratio, similar to those used for interpreting other test results. The work is being done in collaboration with the BayesMendel group at the Bloomberg School of Public Health.

Predicting the impact of somatic variation in cancer genomes with molecular modeling and comparative genomics

The genomes of tumors acquire somatic mutations that may provide insights into their mechanisms of action and potential cancer treatments. For example, the tyrosine kinase inhibitor Iressa has been shown to be effective for non-small-cell lung tumors with missense mutations and small, in-frame deletions near the kinase's ATP-binding pocket. We are developing computational methods based on comparative genomics and protein structure modeling to discriminate between functional and non-functional (passenger) somatic mutants, currently using sequence data from breast and colorectal tumors identified in the Velculescu lab at the Ludwig Center for Cancer Genetics and Therapeutics.

Large scale annotation of human genetic variation

We are developing a genomic-scale software pipeline to annotate human germline variants. The pipeine comprehensively maps SNPs onto protein sequences, functional pathways, and comparative protein structure models. The current output of the pipeline is a set of positions where SNPs are predicted to change an amino acid and thus destabilize the gene's protein product, interfere with the formation of protein domain-domain interfaces, or have an effect on protein-ligand binding. Predicted disease-associations are generated with a supervised machine learning algorithm called a support vector machine, trained on a set of known deleterious mutations from OMIM and a set of putatively benign polymorphisms from dbSNP. The results can be used to identify candidate functional SNPs within a single gene, set of genes, genomic region, or KEGG-defined biochemical pathway.

Molecular modeling of point mutations in proteins

We are developing methods to model the impact of amino acid substitutions on protein structure using molecular dynamics and the MODELLER software package, in collaboration with the Sali lab at UCSF. Proteins are marginally stable under physiological conditions and a changed amino acid can introduce strain into the structural environment. In many cases, the local environment adjusts to accommodate the new amino acid. Otherwise, the protein may be locally or globally destabilized. We are working on a scoring function tailored to missense mutants and benchmarking our simulations against a set of point mutant high resolution x-ray crystal structures.