Quantifying Pathogen Surveillance Using Temporal Genomic Data

Chan, Joseph M.; Rabadan, Raul

Phylogeny as a measurement of surveillance. Sequences were analyzed by using BEAST v1.7.4, a Bayesian MCMC approach, to sample time-structured evolutionary trees from their joint posterior probability distribution. Because of computational and time limits, we restricted our phylogenetic analysis to human H5N1, as BEAST analysis of data sets of over a thousand sequences can take weeks to converge to a stationary condition. This data set was analyzed by using an exponential-growth coalescent as a tree prior with an HKY ϩ ⌫ model of nucleotide substitution. Relaxed and strict molecular clocks were employed. For the strict (relaxed) clock, 20 independent runs of 750,000 (1,500,000) steps each were performed, compared for convergence, and combined, with a burn-in of 150,000 (500,000) steps from each. Each run finished after an average of 32.47 (19.14) h using one 3.00-GHz Intel Xeon CPU with 12 GB of random-access memory. A single maximum clade credibility (MCC) tree was created from each set of simulations with the average length determined for each branch. We found the corresponding evolutionary rates to be 5.11 ϫ 10 Ϫ3 (range, 4.58 ϫ 10 Ϫ3 to 5.62 ϫ 10 Ϫ3 ) and 5.39 ϫ 10 Ϫ3 (range 4.39 ϫ 10 Ϫ3 to 5.80 ϫ 10 Ϫ3 ), similar to the evolutionary rate we used for influenza virus HA analysis. The 95% HPD of each branch length of the MCC tree was used to set a confidence interval for each p2 coefficient.