Presentation at ESCAIDE 2016 by Thibaut Jombart. The R Epidemics Consortium: Building the next generation of statistical tools for outbreak response using R
8980367676 Call Girls In Ahmedabad Escort Service Available 24×7 In Ahmedabad
Building the next generation of statistical tools for outbreak response using R
1. Building the next generation of statistical tools for outbreak
response using R
Thibaut Jombart
28th November 2016
Imperial College London
MRC Centre for Outbreak Analysis and Modelling
1
13. Lessons learnt from the Ebola response
Most statistical/modelling tools for situation awareness missing.
4
14. What tools do we need?
Some examples:
• data cleaning: dictionaries, entry matching
• graphics: case incidence in space and time, contact tracing
• parameter estimation: key delays, transmissibility
• estimate / test CFR: gender, health care workers, treaments
eects
• predictions: case incidence, mortality, evaluate interventions
• report: (semi-)automated situation reports
5
20. Hackout 3: a hackathon for emergency outbreak response
Last summer at the rOpenSci headquarters (Berkeley)
7
21. Hackout 3: from ideas to projects to...
report
data
contacts
delay
time
security
siterep
incidence
peak
estimation
gui
integration
transmission
cleaning
outbreaks
reporting
trend
outbreaker
bayesian
number
shinytracing
cleanr
clusters
collectionefficiency
parsing
unit
automation
censoring
compiled
contact
contacttracing
epicontacts
epiinfo
package
repository
reproducible
anonymised
automated
curation
ecdc
ggplot
interface
linelists
mutations
reproduction
series
testing
vhf
bias
dictionary
epiestim
linelist
parallel
parameters
rates
secure
secured
situation
symptoms
userfriendly
cdc
code
encrypted
encryption
epidemics
exposure
fast
genomics
incubation
opensource
period
reliable
tree
continuous
follow
functional
model
rcpp
synchronised
systems
dashboard
distribution
locations
tools
awareness
format
free
tidy
cases
generation
review
8
22. Hackout 3: from ideas to projects to...
report
data
contacts
delay
time
security
siterep
incidence
peak
estimation
gui
integration
transmission
cleaning
outbreaks
reporting
trend
outbreaker
bayesian
number
shinytracing
cleanr
clusters
collectionefficiency
parsing
unit
automation
censoring
compiled
contact
contacttracing
epicontacts
epiinfo
package
repository
reproducible
anonymised
automated
curation
ecdc
ggplot
interface
linelists
mutations
reproduction
series
testing
vhf
bias
dictionary
epiestim
linelist
parallel
parameters
rates
secure
secured
situation
symptoms
userfriendly
cdc
code
encrypted
encryption
epidemics
exposure
fast
genomics
incubation
opensource
period
reliable
tree
continuous
follow
functional
model
rcpp
synchronised
systems
dashboard
distribution
locations
tools
awareness
format
free
tidy
cases
generation
review
How do we keep momentum once the event is over?
8
23. RECON: the R Epidemics Consortium
A taskforce to build a new generation of outbreak response tools in .
www. repidemicsconsortium. org
9
24. In a nutshell
www. repidemicsconsortium. org
• started 6th September 2016
• 46 people (41 members, 5 board)
• 10 countries, 20 institutions
• ∼ 10 new packages coming
• public forum, blog, online resources
10
25. The RECON forum
A platform for discussing epidemics analysis in .
www. repidemicsconsortium. org/ forum
11
26. The RECON forum
A platform for discussing epidemics analysis in .
www. repidemicsconsortium. org/ forum
Join us! 11
27. RECON package: what do we aim for?
• eciency: useful for improving situation awareness in real
time; cutting-edge, computer-ecient statistical methods
12
28. RECON package: what do we aim for?
• eciency: useful for improving situation awareness in real
time; cutting-edge, computer-ecient statistical methods
• reliability: outputs can be trusted; continuous integration,
extensive unit testing, code review, good practices
12
29. RECON package: what do we aim for?
• eciency: useful for improving situation awareness in real
time; cutting-edge, computer-ecient statistical methods
• reliability: outputs can be trusted; continuous integration,
extensive unit testing, code review, good practices
• accessibility: widely available, easy learning curve; extensive
documentation, tutorials, websites, forum
12
32. epicontacts: handling, visualisation and analysis of epidemio-
logical contacts
www. repidemicsconsortium. org/ epicontacts
[release December 2016] 14
33. outbreaker2: inferring who infects whom in an outbreak
Original outbreaker model: timing of symptoms and pathogen
genomes to infer infectors
(Jombart et al, PLoS Comp Biol, 2014)
15
34. outbreaker2: inferring who infects whom in an outbreak
Original outbreaker model: timing of symptoms and pathogen
genomes to infer infectors
(Jombart et al, PLoS Comp Biol, 2014)
Since outbreaker : new models, data, and questions.
15
35. outbreaker2: inferring who infects whom in an outbreak
Original outbreaker model: timing of symptoms and pathogen
genomes to infer infectors
(Jombart et al, PLoS Comp Biol, 2014)
Since outbreaker : new models, data, and questions.
But: methodological niche fragmented.
15
38. Are dierent methods really... dierent?
Dierent models can lead to very similar implementations.
Can we nd a general formulation?
16
39. What do these model look like?
• a, b, c: dierent types of data
• θ: parameters / augmented data
Data are often assumed to be conditionally independent:
p(a, b, c|θ) = p(a|θ)p(b|θ)p(c|θ)
17
40. What do these model look like?
• a, b, c: dierent types of data
• θ: parameters / augmented data
Data are often assumed to be conditionally independent:
p(a, b, c|θ) = p(a|θ)p(b|θ)p(c|θ)
Components can be treated as plugins.
17
41. outbreaker2: a general cauldron for cooking methods
Use-your-own: data type, likelihood, prior, MCMC.
18
42. outbreaker2: a general cauldron for cooking methods
Use-your-own: data type, likelihood, prior, MCMC.
18
43. outbreaker2: a general cauldron for cooking methods
Use-your-own: data type, likelihood, prior, MCMC.
18
44. outbreaker2: a general cauldron for cooking methods
Use-your-own: data type, likelihood, prior, MCMC.
Modularity is key to generalising approaches
18
46. outbreaker2: a general tool for outbreak reconstruction
• modularity: likelihood, priors,
samplers are all modules
19
47. outbreaker2: a general tool for outbreak reconstruction
• modularity: likelihood, priors,
samplers are all modules
• new 'extensions': contact tracing,
spatial structure, new MCMC
19
48. outbreaker2: a general tool for outbreak reconstruction
• modularity: likelihood, priors,
samplers are all modules
• new 'extensions': contact tracing,
spatial structure, new MCMC
• reliability: continuous integration,
extensive unit testing (aiming for
100% coverage)
19
49. outbreaker2: a general tool for outbreak reconstruction
• modularity: likelihood, priors,
samplers are all modules
• new 'extensions': contact tracing,
spatial structure, new MCMC
• reliability: continuous integration,
extensive unit testing (aiming for
100% coverage)
• prettier: plot methods using ggplot2,
interactive networks visualisation
19
50. outbreaker2: a general tool for outbreak reconstruction
• modularity: likelihood, priors,
samplers are all modules
• new 'extensions': contact tracing,
spatial structure, new MCMC
• reliability: continuous integration,
extensive unit testing (aiming for
100% coverage)
• prettier: plot methods using ggplot2,
interactive networks visualisation
• should facilitate new contributions
19
65. Outbreak response context creates distance and delays
• ecient tools can shorten delays
• potential of embedding methodologists in outbreak response
teams 21