This repository provides code for the phylosamp R package, which was designed to help users conduct and evaluate sample size calculations for phylogenetic studies. Presently, the functions can be used to calculate sample size in three types of scenarios that frequently arise when analyzing pathogen genomic data: (1) trying to determine if pathogen infections are linked by transmission (linkage scenario); (2) trying to estimate the frequency of a known pathogen lineage or variant of concern (variant tracking scenario); (3) trying to determine if pathogen transmissibility differs between groups of infected hosts (relative R scenario).
All key functions of each scenario are documented, along with realistic examples, in the associated vignettes. Vignettes are organized as follows:
- Transmission Linkage Vignettes (L1-L4): linkage scenario vignettes and examples
- Variant Tracking Vignettes (V1-V6): variant tracking scenario vignettes and examples
- Differential Transmission Vignettes: relative R scenario vignettes coming soon!
The package includes a suite of functions that can be used to determine the ability of some phylogenetic criteria (e.g., genetic distance) to correctly identify pairs of pathogen infections linked by transmission, given a particular sample size or proportion. The package also includes functions that do the reverse: calculate the sample size needed to correctly identify true pairs at some particular rate.
The functions used to calculate the sample size or false discovery rate of the criteria require as input an estimate of the sensitivity and specificity of the linkage criteria used. Therefore, the current implementation of the package also includes functions to estimate the sensitivity and specificity of genetic distance as a linkage criteria, from the mutation rate of the pathogen. Future implementations may contain guidance on how to estimate these parameters for other phylogenetic criteria.
All functions require the user to specify the underlying assumptions about transmissions and linkage, i.e., if an infected individual can transmit to more than one susceptible individual (single transmission/multiple transmissions), and if the criteria being used is capable of linking a case to more than one other case (single linkage/multiple linkage). Permitting multiple transmissions and multiple links ('mtml') is the default.
A detailed description of the linkage methods can be found in:
Sample Size Calculation for Phylogenetic Case Linkage (Wohl, Giles, and Lessler 2020)
The package includes another set of functions that can be used to determine the sample size needed to detect or estimate the frequency of a pathogen variant in a population. It also includes functions that do the reverse: calculate the confidence in a detection or frequency estimate, given a number of samples.
These functions require the user to specify a desired confidence in the results (either probability of detection or confidence in prevalence estimate), a desired or estimated variant prevalence, and (if applicable) a desired precision in the prevalence estimate. The user can also provide variant-specific parameters to help account for biases in variant detection, such as the probability that an infection is asymptomatic, the testing sensitivity, the testing probability, and the sequencing success rate.
Functions are provided for sample size calculations in a cross-sectional scenario, where a single batch of samples will be collected and sequenced, and in a periodic surveillance scenario, where samples are collected and sequenced repeatedly at some regular interval. In the latter case, the user must also provide information on how the variant frequency may be changing over time (in the form of an initial frequency and estimated logistic growth rate).
All calculations assume a two-variant system; in other words, that there is a particular variant of interest that may behave differently from the rest of the pathogen population (in terms of asymptomatic rate, testing sensitivity, etc.). Detection biases due to these differences are incorporated into function calculations. That said, the framework could be easily extended to a multi-variant system in the future.
A detailed description of the VOC estimation methods can be found in:
Finally, the package includes an additional set of functions that can be used to determine the sample size needed to detect differential transmission between groups of potential pathogen hosts. It also includes functions that do the reverse: calculate the power given parameters such as the effect size and the number of samples.
We assume the user is interested in detecting differential transmission between two groups of individuals, denoted as A and B, which can optionally be of different sizes. The main function allows the user to specify the estimated reproductive number in group A, the estimated reproductive number in group B, the proportion of the infected population that are in group A (equivalent to 1 minus the proportion in group B), the total size of the outbreak, the desired type 1 error rate, and the desired power. These parameters are then used to calculate the sample size needed to detect if the reproductive number differs between groups A and B.
The user can specify whether they are interested in a one-sided or two-sided hypothesis test and whether there is any linkage misclassification, via sensitivity and specificity parameters. Additionally, the user can optionally specify an overdispersion parameter if overdispersion is suspected in the transmission process.
A detailed description of the differential transmission methods can be found in:
The phylosamp package is available for download on CRAN.
To install the install the development version of the phylosamp package, first install the devtools package and then install phylosamp from source via GitHub:
install.packages('devtools')
devtools::install_github('HopkinsIDD/phylosamp')This package is maintained by Elizabeth Lee (@eclee25), Shirlee Wohl (@shwohl), and Justin Lessler (@jlessler).
For general questions, contact Shirlee Wohl ([email protected]), or Justin Lessler ([email protected]).
To report bugs or problems with documentation, please go to the Issues page associated with this GitHub page and click new issue.