Christopher Ryan King
University of Chicago
I am a trainee of the Medical Scientist Training Program (MSTP) dual MD/PhD program in the Pritzker School of Medicine at the University of Chicago.
My PhD in biostatistics was completed in the Health Studies department.
I defended my dissertation May 2012.
I'm applying for anesthesiology residency spots this year.
My PhD focused on statistical genetics, in particular methods for phenotype association in resequencing studies.
There are a few questions which made up my dissertation:
There's also a lot of implementation that went with that. I wrote and modified software to
- How methods or hypotheses which look for larger effects among rare variants are related to hypotheses about the evolution of the phenotype.
- How observing more rare variants in cases in a case-control study does not imply that rare variants are on average disease promoting.
- How studies which follow up on SNPs discovered via sequencing in case-control studies can observe an exaggerated difference between rare and common SNPs, and how to fix estimates up to account for ascertainment of cases.
- How to predict risk for binary traits for individuals with novel SNPs in associated genes.
- Some estimation details and identification issues in semi-parametric models of the SNP-minor allele frequency relationship in case-control studies.
The datasets I ended up working with are
- Estimate and test association models for quantitative traits using a random effects approach.
- Estimate and test association models for binary traits using MCMC.
- Estimate the distribution of minor allele frequencies in the source population of a case-control study.
- Use the above to get a corrected and stable estimate of the effect-frequency relationship in a case-control study.
There are a number of extensions which are in an advanced stage of progress but didn't make it into the dissertation:
- The Dallas Heart Study resequencing of several lipids-associated genes.
- A case-control resequencing of asthmatic individuals provided by the Ober lab at U of C.
- Whole-genome sequencing of Mexican-American families as part of the T2D-GENES project.
There are some side projects that I may never actually do, but am keeping on the back-burner for future research:
- A method for family-based sequencing which estimates the correlation between robust within-family SNP effect estimates and powerful across-family estimates. That correlation allows
This method is different from and more robust that previous work attempting to combine distinct studies with different designs.
- Estimation of the degree of residual confounding genome-wide and stratified by SNP type.
- Adaptive pooling of between and within family information. Early testing indicates that it increases the power of the test vs no effect when confounding is low and does not over-pool when confounding is high.
- Evaluation of the widely applicable information criterion (WAIC) as a test statistic for multi-level models using the genome-wide set of sequence matrices and simulation-based p-values. Comparison vs Bayesian Wald-type statistics and REML LR and Wald tests in gaussian models identify substantial differences on certain design matrices
- Adaptation of Wang-Landau algorithms (aka SAMC) for high-throughput evaluation of expensive test statistics in relatively short runs. Initialization is accomplished by the genome-wide distribution of test statistics and small number of independent simulations. Smoothing and update parameters are automatically determination from observed chain diffusion and effective sample size. Procedure termination is determined by observed estimate stability. Early experiments show about a 10-fold decrease in number of test-statistic evaluations to estimate p-values about 1e-6 versus independent simulation. Currently working to replace the Gibbs update step in SAMC with a Hamiltonian one (Gibbs-ing into very rare events requires slow diffusion and lots of iterations).
- Novel updates inside of SAMC for family-based data: partial gene-dropping and propensity score resampling of founder haplotypes.
- A comparison of methods for robust-to-confounding association testing and estimation in family-based sequencing:
- Mixed conditional logistic regression making an invalid multiplicative model assumption
- A heuristic reweighing to conditional logistic estimates which should be valid under the null.
- A control-function / adjusted instrumental variable approach.
- A nearly-saturated adjustment for pedigree membership and poly-gene residual correlation.
- Variational-Bayes techniques to estimate association in family-based data as 'adjusted' unrelateds. This ended up not working well and put on hold, but could be worth pursuing.
- A weighted combination of Laplace approximation and pairwise pseudo-likelihood (equivalent to some variational procedures) for rapid evaluation of the likelihood for SNP-based GLMMs. The laplace approximation is excellent for integrating out effects of SNPs which are common and therefore have a fair amount of information, but terrible for SNPs which are rare in the sample.
On the other hand, pairwise (or other approximate factorizations of the likelihood) is great for effects which only show up a small number of times but inefficient and slow for common SNPs. This ended up not working well in our dataset, but could be improved if I change the variational part.
My primary adviser was Dan Nicolae in the Departments of Statistics, Human Genetics, and Medicine. My committee chair was Paul Rathouz who recently left the Department of Health Studies for Biostatistics at UW-Madison.
Jonathan Pritchard (Human Genetics), Ronald Thisted (Health Studies), and Dezheng Huo (Health Studies) are also on my committee.
I previously did some applied sleep / cardiovascular epidemiology with Diane Lauderdale, also in Health Studies at the University of Chicago.
Before coming to Chicago, I worked with Boris Kovatchev at the University of Virginia on methods for continuous glucose sensors.
- King CR, Rathouz PJ, Nicolae DL, 2010 An Evolutionary Framework for Association Testing in Resequencing Studies. PLoS Genet 6(11): e1001202. doi:10.1371/journal.pgen.1001202 Supplementary Code.
- King CR, Knutson KL, Rathouz PJ, Sidney S, Liu, K, Lauderdale DS. Short sleep duration and incident coronary artery calcification. JAMA. 2008;300(24):2859–2866.
- Kovatchev, B.P. King C, Breton M, Anderson S, Clarke W. Clinical assessment and mathematical modeling of the accuracy of continuous glucose sensors (CGS). Conf Proc IEEE Eng Med Biol Soc 1, 71-4(2006).
- King CR, Anderson SM, Breton MD, Clarke WL, Kovatchev BP: Modeling of calibration effectiveness and blood-to-interstitial glucose dynamics as potential confounders of the accuracy of continuous glucose sensors during hyperinsulinemic clamp. J Diabetes Sci Technol 1:317–322, 2007
I received a PhD in Health Studies from the University of Chicago in 2012.
I received a BS in Physics and Mathematics from the University of Virginia in 2005.
I use email@example.com.
This is what I looked like in 2007