Assessing accuracy and precision in phylodynamic reconstruction


Trevor Bedford (@trvrb)
25 Mar 2019
VEG Meeting
Fred Hutch

Virus genomes reveal factors that spread and sustained the Ebola epidemic

with Gytis Dudas, Andrew Rambaut, Luiz Carvalho, Marc Suchard, Philippe Lemey,
and many others

Dudas et al. 2017. Nature.

Sequencing of 1610 Ebola virus genomes collected during the 2013-2016 West African epidemic

Dudas et al. 2017. Nature.

Phylogenetic reconstruction of epidemic

Dudas et al. 2017. Nature.

Tracking migration events

Dudas et al. 2017. Nature.

Factors influencing migration rates

Dudas et al. 2017. Nature.

Mixing at the regional level

Dudas et al. 2017. Nature.

Each introduction results in a minor outbreak

Dudas et al. 2017. Nature.

The ability of single genes vs full genomes to resolve time and space in outbreak analysis

with Gytis Dudas

Dudas and Bedford. 2019. bioRxiv.

Accuracy vs precision

Schoene et al. 2013. Elements.

Approach accuracy and precision through a machine learning approach to model testing. Leave out 60/600 tips and predict time and location of these tips. Compare:

  • Single gene vs full genome
  • BEAST vs TreeTime

Maximum likelihood divergence trees

BEAST time trees

Date reconstruction

Evolutionary rate reconstruction

Location reconstruction

Estimates are generally well calibrated

Correlates of migration via GLM

Tracing migration histories

Tracing migration histories

Date reconstruction via TreeTime

Location reconstruction via TreeTime

GP vs genome:

  • GP date error of 106 days vs genome date error of 22 days
  • GP location accuracy of 29% vs genome location accuracy of 54%
  • GP location error of 164 km vs genome location error of 76 km

BEAST vs TreeTime for full-genome:

  • BEAST date precision of 22 days vs TreeTime date error of 29 days
  • BEAST location accuracy of 54% vs TreeTime location accuracy of 43%
  • BEAST location error of 76 km vs TreeTime location error of 87 km

Theoretical resolution based on evolutionary rate and genome size

Theoretical resolution based on evolutionary rate and genome size

Acknowledgements

Bedford Lab: Alli Black, John Huddleston, Barney Potter, James Hadfield,
Louise Moncla, Tom Sibley, Maya Lewinsohn, Katie Kistler

Ebola: Gytis Dudas, Andrew Rambaut, Luiz Carvalho, Philippe Lemey, Marc Suchard, Andrew Tatem

Phylogenetic resolution: Gytis Dudas