Research Interests for Oliver Ratmann

Research Interests: Virus ecology and evolution, Network evolution, Bayesian modeling, Bayesian model choice and model criticism.

Approximate Bayesian Computation

Approximate Bayesian Computation (ABC) is a developing area of statistics that is motivated by the need to fit, assess and compare models that have a well-defined underlying generative process, but for which classical statistical procedures of estimation and goodness-of-fit encounter formidable computational challenges. ABC only requires to simulate efficiently from a mathematical model. It circumvents likelihood calculations by comparing simulations from a model with all the data in terms of many summaries at the same time, and embeds these comparisons into the Bayesian framework for the purpose of parameter inference.

My doctoral thesis is on ABC under model uncertainty (ABCmu), an extension to ABC that makes possible to analyze goodness-of-fit of these models at no extra computational cost.

Here are some comments on this work
  - ABC for model criticism
  - Discussion of ABC for model criticism
  - Doctoral thesis
I would like to thank my collaborators, Profs. Christophe Andrieu (Bristol, UK), Carsten Wiuf (BiRC, DK) and my PhD supervisor Prof. Sylvia Richardson (Imperial College London, UK) for their continued, stimulating advice.

Evolutionary analysis of protein interaction networks

The evolutionary mechanisms by which protein interaction networks grow and change are beginning to be appreciated as a major factor shaping their present-day structures and properties. Since the first observations of unexpected structure in the yeast protein interaction data, focus has shifted somewhat from the description of large-scale topological features via simple models of how such features may have evolved toward a broader and more subtle appreciation of the underlying biological mechanisms of network evolution and the effects of sampling, bias, and experimental uncertainty on the available data.

Our quantitative analyses in the Bayesian framework support the view that the nonrandom topological features of protein interaction networks may have simply been determined by evolutionary genetic mechanisms such as gene duplication or point mutations. Specifically, models that caricature molecular genetic "copy-and-paste" mechanisms lead to local network structures and degree correlations that are better supported by the data than models based on various forms of preferential attachment and link turnover. We also found that a comprehensive set of summary statistics is required for reliable and consistent inference, and that the node degree distribution alone is not sufficient to reflect the complexities of the network topology. Finally, evolutionary interpretations may be extremely fragile under different assumptions about network incompleteness, suggesting that, for the analysis of binary interaction data, the various forms of measurement error need to be carefully accounted for.

Many thanks to Prof. Carsten Wiuf (BiRC, DK) and his research group, and Trevor Hinkley (ETH, Zuerich, CH) for their help and support.

Unraveling the dynamics of rapidly evolving infectious diseases in humans.

With the world-wide spread of HIV and the re-emergence of “eradicated” pathogens, faith in the effectiveness of public health programs has been dwindling.

One crucial problem of efforts targeted at improving public health is that many viral pathogens are evolving rapidly, thus re-juvenating persistently, and turning drugs and vaccines quickly ineffective. Many variants of these viruses coexist at any time, with each variant “challenged” by different levels of host immunity. As time progresses, new mutations and the history of pre-existing immunity in the population render the resulting virus dynamics extremely complicated.

Recent technological advances are now delivering data on the immunogenic properties, the genetic diversity and the ancestral relationship of these variants, which together with case data hold the potential to unravel the dynamics of rapidly evolving infectious diseases. But these data are bewilderingly heterogeneous, making it difficult to write down a formula for the probability of all the data (the “likelihood”) under specific assumptions on the nature of these dynamics.

I am currently adapting methods of ABC to study quantitatively and rigorously the ecological, immunogenic and evolutionary processes that shape the dynamics or rapidly evolving infectious diseases. Many thanks to my supervisor Dr. Katia Koelle (Duke, USA) and my collaborator Prof. Christophe Fraser (Imperial College London, UK).

Representative Publications   (search)
  1. Ratmann O, Andrieu C, Wiuf C and Richardson S, Reply to Robert et al.: Model criticism informs model choice and model comparison, Proc Natl Acad Sci USA, vol. 107 no. 3 (January, 2010), pp. E6-E7 [doi[abs]
  2. Ratmann O, Andrieu C, Wiuf C and Richardson S, Model criticism based on likelihood-free inference, with an application to protein network evolution, Proc Natl Acad Sci USA, vol. 106 no. 26 (June, 2009), pp. 10576-10581 [doi[abs]
  3. Ratmann O, Wiuf C, Pinney JW, From evidence to inference: probing the evolution of protein interaction networks, HFSP Journal, vol. 3 no. 5 (December, 2009), pp. 290-306 [doi[abs]
  4. Ratmann O, Jorgensen O, Hinkley T, Stumpf MPH, Richardson S and Wiuf C, Using Likelihood-Free Inference to Compare Evolutionary Dynamics of the Protein Networks of H. pylori and P. falciparum, PLoS Comp Biol, vol. 3 no. 2007 (November, 2007), pp. e230 [doi[abs]