Limits...
Measuring Asymmetry in Time-Stamped Phylogenies.

Dearlove BL, Frost SD - PLoS Comput. Biol. (2015)

Bottom Line: We present a simple permutation-based approach for testing for asymmetry in a phylogeny, where we compare the observed phylogeny with random phylogenies with the same sampling and coalescence times, to reduce the false positive rate.We also demonstrate how profiles of measures of asymmetry calculated over a range of evolutionary times in the phylogeny can be used to identify local asymmetry.In combination with different metrics of asymmetry, this combined approach offers detailed insights of how phylogenies reconstructed from real viral datasets may deviate from the simplistic assumptions of commonly used coalescent and birth-death process models.

View Article: PubMed Central - PubMed

Affiliation: Department of Veterinary Medicine, University of Cambridge, Cambridge, United Kingdom.

ABSTRACT
Previous work has shown that asymmetry in viral phylogenies may be indicative of heterogeneity in transmission, for example due to acute HIV infection or the presence of 'core groups' with higher contact rates. Hence, evidence of asymmetry may provide clues to underlying population structure, even when direct information on, for example, stage of infection or contact rates, are missing. However, current tests of phylogenetic asymmetry (a) suffer from false positives when the tips of the phylogeny are sampled at different times and (b) only test for global asymmetry, and hence suffer from false negatives when asymmetry is localised to part of a phylogeny. We present a simple permutation-based approach for testing for asymmetry in a phylogeny, where we compare the observed phylogeny with random phylogenies with the same sampling and coalescence times, to reduce the false positive rate. We also demonstrate how profiles of measures of asymmetry calculated over a range of evolutionary times in the phylogeny can be used to identify local asymmetry. In combination with different metrics of asymmetry, this combined approach offers detailed insights of how phylogenies reconstructed from real viral datasets may deviate from the simplistic assumptions of commonly used coalescent and birth-death process models.

No MeSH data available.


Related in: MedlinePlus

Permutations of an observed tree can overcome bias in detecting asymmetry in time-sampled phylogenies.a) An ‘observed’ tree, simulated under the coalescent model with 100 sequences sampled over 10 time points, each 1000 generations apart, with effective population size of 104. b) The distribution of Sackin’s index and number of cherries for 100 random trees, simulated as in a) except for tips being sampled at a single time point. Expected values for these distributions are shown with dashed black lines. The observed values (solid black line) are highly extreme due to the implicit bias caused by tips sampled early in the ancestry. However, this is not the case when comparing them to a distribution calculated from permuting the observed tree, as seen in c), where there is no evidence to suggest the observed tree is asymmetric and the solid black line falls between the 2.5% and 97.5% quantiles (dashed red lines).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4492995&req=5

pcbi.1004312.g003: Permutations of an observed tree can overcome bias in detecting asymmetry in time-sampled phylogenies.a) An ‘observed’ tree, simulated under the coalescent model with 100 sequences sampled over 10 time points, each 1000 generations apart, with effective population size of 104. b) The distribution of Sackin’s index and number of cherries for 100 random trees, simulated as in a) except for tips being sampled at a single time point. Expected values for these distributions are shown with dashed black lines. The observed values (solid black line) are highly extreme due to the implicit bias caused by tips sampled early in the ancestry. However, this is not the case when comparing them to a distribution calculated from permuting the observed tree, as seen in c), where there is no evidence to suggest the observed tree is asymmetric and the solid black line falls between the 2.5% and 97.5% quantiles (dashed red lines).

Mentions: As an example, consider a tree simulated with 100 tips sampled over 10 time points (Fig 3a). Comparing this heterochronous tree with 1000 similarly simulated trees but with tips sampled at a single time point (Fig 3b) illustrates how extreme the observed values of Sackin’s index and number of cherries (solid black line) are compared to the expected values (dashed black line), purely due to the serial sampling [5]. However, when the heterochronous observed tree is compared to a distribution obtained from the permuted trees (Fig 3c), it can be seen that in the distribution of possible trees with the same internal node and tip sampling times, there is little evidence to suggest that this observed tree is asymmetric.


Measuring Asymmetry in Time-Stamped Phylogenies.

Dearlove BL, Frost SD - PLoS Comput. Biol. (2015)

Permutations of an observed tree can overcome bias in detecting asymmetry in time-sampled phylogenies.a) An ‘observed’ tree, simulated under the coalescent model with 100 sequences sampled over 10 time points, each 1000 generations apart, with effective population size of 104. b) The distribution of Sackin’s index and number of cherries for 100 random trees, simulated as in a) except for tips being sampled at a single time point. Expected values for these distributions are shown with dashed black lines. The observed values (solid black line) are highly extreme due to the implicit bias caused by tips sampled early in the ancestry. However, this is not the case when comparing them to a distribution calculated from permuting the observed tree, as seen in c), where there is no evidence to suggest the observed tree is asymmetric and the solid black line falls between the 2.5% and 97.5% quantiles (dashed red lines).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4492995&req=5

pcbi.1004312.g003: Permutations of an observed tree can overcome bias in detecting asymmetry in time-sampled phylogenies.a) An ‘observed’ tree, simulated under the coalescent model with 100 sequences sampled over 10 time points, each 1000 generations apart, with effective population size of 104. b) The distribution of Sackin’s index and number of cherries for 100 random trees, simulated as in a) except for tips being sampled at a single time point. Expected values for these distributions are shown with dashed black lines. The observed values (solid black line) are highly extreme due to the implicit bias caused by tips sampled early in the ancestry. However, this is not the case when comparing them to a distribution calculated from permuting the observed tree, as seen in c), where there is no evidence to suggest the observed tree is asymmetric and the solid black line falls between the 2.5% and 97.5% quantiles (dashed red lines).
Mentions: As an example, consider a tree simulated with 100 tips sampled over 10 time points (Fig 3a). Comparing this heterochronous tree with 1000 similarly simulated trees but with tips sampled at a single time point (Fig 3b) illustrates how extreme the observed values of Sackin’s index and number of cherries (solid black line) are compared to the expected values (dashed black line), purely due to the serial sampling [5]. However, when the heterochronous observed tree is compared to a distribution obtained from the permuted trees (Fig 3c), it can be seen that in the distribution of possible trees with the same internal node and tip sampling times, there is little evidence to suggest that this observed tree is asymmetric.

Bottom Line: We present a simple permutation-based approach for testing for asymmetry in a phylogeny, where we compare the observed phylogeny with random phylogenies with the same sampling and coalescence times, to reduce the false positive rate.We also demonstrate how profiles of measures of asymmetry calculated over a range of evolutionary times in the phylogeny can be used to identify local asymmetry.In combination with different metrics of asymmetry, this combined approach offers detailed insights of how phylogenies reconstructed from real viral datasets may deviate from the simplistic assumptions of commonly used coalescent and birth-death process models.

View Article: PubMed Central - PubMed

Affiliation: Department of Veterinary Medicine, University of Cambridge, Cambridge, United Kingdom.

ABSTRACT
Previous work has shown that asymmetry in viral phylogenies may be indicative of heterogeneity in transmission, for example due to acute HIV infection or the presence of 'core groups' with higher contact rates. Hence, evidence of asymmetry may provide clues to underlying population structure, even when direct information on, for example, stage of infection or contact rates, are missing. However, current tests of phylogenetic asymmetry (a) suffer from false positives when the tips of the phylogeny are sampled at different times and (b) only test for global asymmetry, and hence suffer from false negatives when asymmetry is localised to part of a phylogeny. We present a simple permutation-based approach for testing for asymmetry in a phylogeny, where we compare the observed phylogeny with random phylogenies with the same sampling and coalescence times, to reduce the false positive rate. We also demonstrate how profiles of measures of asymmetry calculated over a range of evolutionary times in the phylogeny can be used to identify local asymmetry. In combination with different metrics of asymmetry, this combined approach offers detailed insights of how phylogenies reconstructed from real viral datasets may deviate from the simplistic assumptions of commonly used coalescent and birth-death process models.

No MeSH data available.


Related in: MedlinePlus