Tuesday - November 30, 2004

*The location of this lecture has been changed.  It will now be held in Room DE at the Radisson Hotel RTP.  Click here for directions.  

HIV-1 is a highly variable pathogen, and the variation is in part the direct the result of immune escape. Every person carries a population of distinct of the virus; generally viruses sampled from early in the infection are very similar, and within-patient diversity builds over time. HIV has in its repertoire many evolutionary strategies that increase the diversity, not only base substitution, but recombination, frequent insertion and deletions, and changes in patterns of glycosylation. Population diversity grows over time, evolving outward within sets of distinct lineages called subtypes. Inter-subtype recombination events are common, and such recombinants can found their own lineages. Part of the difficult in making an effective vaccine is overcoming this diversity. We are working at Los Alamos on the task of designing vaccine antigens that have increased potential to stimulate cross-reactive responses. Our strategies to date involves using artificial consensus sequences and maximum-likelihood derived ancestral sequences to find an approximation of a central position in antigenic space, and to derive an artificial antigen that has better cross-reactive potential with circulating strains than any single natural strain. Our experimentalist collaborators at Duke and University of Alabama have tested these constructs, and the initial results are promising. We are now turning our attention to focusing on sequences derived from acute infection, as at least in the case of A and C subtypes, these seem to have shared distinctive properties and may have conserved features that make them a narrower target than the full spectrum of viruses isolated at any stage of progression. We are also developing strategies to define combinations of proteins that could provide maximum coverage of a population. 

Tuesday - January 25, 2005

*Note:  This lecture will be held in Room H at the Radisson Hotel RTP

Eugenia Kalnay
University of Maryland, Department of Meteorology
Data Assimilation and Ensemble Forecasting:

Two Problems with the Same Solution?

Until 1991, operational numerical weather prediction models utilized a single control forecast representing the best estimate of the state of the atmosphere at the initial time. In 1992, operational NWP models began to utilize ensembles of forecasts from slightly perturbed initial conditions. Such ensemble forecasts provide human forecasters with a range of possible solutions, whose average is generally more accurate than the single deterministic forecast, and whose spread gives information about the forecast errors. It also provides a quantitative basis for probabilistic forecasting.
The two essential problems in the design of an ensemble forecasting system are how to create effective initial perturbations, and how to handle model deficiencies, which make the ensemble forecast spread smaller than the forecast error. In this talk we present a brief historic review of ensemble forecasting and current methods to create perturbations. We point out that the promising approach of ensemble Square Root Kalman Filtering for data assimilation can solve, at the same time, the problems of obtaining optimal initial ensemble perturbations, and possibly estimating the impact of model errors. We also discuss the problem of coupled systems with instabilities that have very different time scales. 

Monday - January 31, 2005

Tuesday - March 29, 2005