Transcript NEA

Grid enabling phylogenetic inference on
virus sequences using BEAST - a
possibility?
EUAsiaGrid Workshop
4-6 May 2010
Chanditha Hapuarachchi
Environmental Health Institute
National Environment Agency
Outline
 Work scope
 Analytical approach
 Current limitations
 What is expected from Grid-enabling?
Work scope
 Understanding the molecular epidemiology of vector-borne, infectious
diseases in Singapore with a view of utilizing information in disease control
operations
Objectives
 To determine the routes of pathogen migration (mainly Dengue and
Chikungunya viruses)
 To understand the evolutionary dynamics of pathogens
 To understand the outbreak potential of pathogens within the country
What phylogenetic inferences are made?
Temporo-spatial
distribution of viruses
Phylogenetic relationships
(BEAST, NETWORK)
(BEAST, MEGA)
(trees)
Molecular
epidemiology
of DENV &
CHIKV
Population dynamics
(Bayesian skyline plots)
(BEAST)
Evolutionary dynamics
(Evolutionary rates, selection
pressure, recombination etc)
(BEAST, HYPHY etc.)
BEAST is a multi-task software package
CHIKV whole genome tree with spatial model
India
Sri Lanka
Singapore
Malaysia
Ind. Ocean Islands
Kenya
Time (yrs)
Spatial distribution of different lineages of DENV in Singapore
However……..
BEAST analysis is time consuming & requires
substantial computing power
Limitations of the BEAST approach?
 Size of dataset
Length of sequences
No. of sequences
E.g. Analyzing a dataset of ~90 whole genomes of CHIKV (11.8 kb)
takes several days depending on the available computing power
Limitations…
 Analytical parameters
 A basic analysis takes ~0.3 hrs per million states
(Core 2 duo, 2.1 GHz, 4 GB RAM, >50% CPU)
 A general run involves at least a 100 million sampling frame
(=~30 hrs)
 The duration increases substantially with changing
parameters
Incorporation of spatial model (7 states) alone
increases the runtime to ~0.4 hrs per million states
 The ultimate duration depends on Effective Sample Size (ESS)
values (general requirement >200)
BEAST Tracer output window
Limitations…
 Number of parallel runs & users
↑ runs & users -------- ↓ analytical efficiency
Single run takes up >50% of CPU power
Why to Grid-enable BEAST?
 Enables efficient data analysis
parallel runs
multiple users
expanded datasets
 Enhances data interpretation
Can Grid-enabling help to improve the existing
performance?