rSalvador: An R Package for the Fluctuation Experiment

Abstract

The past few years have seen a surge of novel applications of the Luria-Delbrück fluctuation assay protocol in bacterial research. Appropriate analysis of fluctuation assay data often requires computational methods that are unavailable in the popular web tool FALCOR. This paper introduces an R package named rSalvador to bring improvements to the field. The paper focuses on rSalvador’s capabilities to alleviate three kinds of problems found in recent investigations: (i) resorting to partial plating without properly accounting for the effects of partial plating; (ii) conducting attendant fitness assays without incorporating mutants’ relative fitness in subsequent data analysis; and (iii) comparing mutation rates using methods that are in general inapplicable to fluctuation assay data. In addition, the paper touches on rSalvador’s capabilities to estimate sample size and the difficulties related to parameter nonidentifiability.

Scalable, Continuous Evolution of Genes at Mutation Rates above Genomic Error Thresholds

“…To calculate per-base substitution rates, fluctuation data were analyzed by the maximum likelihood method, implemented using newton.LD.plating in rSalvador 1.7 (Zheng, 2017). Phenotypic mutation rates were calculated by normalizing to the average number of cells per culture.…”

Bacterial defenses against a natural antibiotic promote collateral resilience to clinical antibiotics

“…This is a maximum likelihood-based method for inferring mutation rates from fluctuation test colony counts, based on the classic Luria-Delbrück (LD) distribution with a correction to account for the effects of partial plating (i.e. plating a portion of each culture rather than the total volume) 77 . We chose this method because it has been shown to be the most accurate estimator of m when partial plating is involved 77,78 .…”

“…plating a portion of each culture rather than the total volume) 77 . We chose this method because it has been shown to be the most accurate estimator of m when partial plating is involved 77,78 . To get µapp (apparent mutation rate per generation) from m, we divided m by the total number of cells per parallel culture 77 , as estimated from the mean number of CFUs counted for the six representative cultures.…”

“…To compare the Hamon and Ycart model to the mixed LD-Poisson model, we simply compared the negative loglikelihoods (smaller values indicate a better fit); the likelihood ratio test was not applicable as these two models contain the same number of parameters. Note that although the Hamon and Ycart (or in some cases, LD-Poisson) models were often better fits than the basic LD model, we still used the basic LD model for statistical comparison of mutation rates between conditions, because an accurate method to account for partial plating has not yet been developed for the cases of post-plating mutations or differential fitness between mutants and parent strains 77 .…”

Evolutionary epidemiology ofStreptococcus iniae: linking mutation rate dynamics with adaptation to novel immunological landscapes

“…Likelihood Ratio Test (LRT) using the rSalvador R package (Zheng 2015(Zheng , 2017. LRT estimates the 600 overlap between confidence intervals calculated for two fluctuation experiment data sets and allows 601 to compare fluctuation assay data with different number of cultures and resolve very small differences 602 in mutation rate (Zheng 2015(Zheng , 2017.…”