A Bayesian model for estimating with-in host P. falciparum haplotype frequencies

Aimee R. Taylor, Jennifer A. Flegg, Philippe J. Guérin, Cally Roper, Chris HolmesMalaria 2012 · DOI

Abstract

Effective surveillance of antimalarial resistance is increasingly critical due to the emergence of resistance to artemisinin derivatives. However, identifying resistant haplotypes in patient samples is complicated by Multiplicity of Infection (MOI), where multiple parasite clones coexist within a single host, leading to convoluted genetic data. We present a Bayesian model designed to reconstruct these hidden haplotypes and estimate their frequencies from prevalence data without requiring prior knowledge of MOI. Using a Metropolis-Hastings Markov Chain Monte Carlo (MCMC) algorithm, the model explores possible haplotype compositions and provides frequency distributions with credible intervals. Validated against simulated datasets and applied to historic African data on sulphadoxine-pyrimethamine resistance, the model demonstrates high accuracy and utility for public health surveillance.

Addressing the Challenge of Multiplicity of Infection

The rise of drug-resistant Plasmodium falciparum poses a significant threat to global malaria control efforts. To track the spread of resistance, researchers must identify specific genetic mutations (haplotypes) associated with drug failure. However, in many endemic regions, patients are frequently infected with more than one parasite strain simultaneously.

The biological complexity of malaria infections often involves multiple parasite clones within a single host.

When a blood sample from such a patient is genotyped, the resulting data is "convoluted." For instance, if multiple mutations are detected at different loci, it is impossible to tell which mutations belong to which specific parasite clone using standard prevalence data. This ambiguity makes it difficult to accurately measure the frequency of highly resistant strains within a population.

Methodology: The Bayesian Model

To solve this "genetic soup" problem, we developed a Bayesian model that uses statistical inference to de-convolve individual haplotypes from population-level prevalence data.

The MCMC Algorithm

The core of the model is a Metropolis-Hastings Markov Chain Monte Carlo (MCMC) algorithm. This stochastic process systematically explores the vast space of possible haplotype combinations that could explain the observed prevalence data. For each iteration, the model calculates the likelihood of the data based on its current estimate of global haplotype frequencies.

Key Advantages

  • No Prior MOI Required: Unlike many previous methods, this model does not need to know the number of infecting strains (MOI) for each patient beforehand—a variable that is often unknown or poorly estimated in field studies.
  • Probabilistic Output: Rather than providing a single point estimate, the model generates a distribution of frequency estimates. This allows for the calculation of the mean frequency and a credible interval, giving researchers a clear measure of statistical certainty.

Validation and Real-World Application

The model's performance was rigorously tested using two distinct approaches:

  1. Simulated Data: By testing the model on datasets where the "ground truth" (the actual haplotypes and their frequencies) was known, we confirmed that the algorithm could accurately recover the true genetic landscape from convoluted samples.
  2. Historic African Data: We applied the model to retrospective data from Africa concerning mutations linked to sulphadoxine-pyrimethamine (SP) resistance. The model successfully estimated the prevalence of specific resistant haplotypes, demonstrating its practical value for analyzing long-term trends in drug resistance.

Conclusion and Future Directions

This Bayesian approach provides a robust tool for public health officials to monitor antimalarial resistance more accurately. By providing credible intervals, it offers a nuanced view of the "risk bounds" for resistance in a given region. While currently validated on SP resistance markers, the model is designed to be adaptable to other molecular markers, including the kelch13 gene associated with artemisinin resistance. Future work will focus on optimizing the computational efficiency of the MCMC algorithm to handle larger datasets and more complex multi-locus surveillance.

Summary updated

Abstract

The necessity for effective surveillance of antimalarial resistance is all the more reinforced by the recent emergence of resistance to artemisinin derivatives. However, the accurate detection of resistant haplotypes from patient samples that are infected with multiple parasite clones is nontrivial. When the multiplicity of infection exceeds one, the allele sequences of the constituent clones at genotyped loci are convoluted. Nevertheless, statistical methods can be used to reconstruct the allele sequences, infer distinct haplotypes and ascertain their frequencies from prevalence data. We have developed a Bayesian model for estimating haplotype frequencies. The model estimates haplotype frequencies based on prevalence data collected for one or more molecular markers known to been associated with antimalarial resistance. Prior knowledge of the MOI is not required. The model uses a Metropolis-Hasting Monte Carlo Markov chain algorithm to explore the different possible haplotype compositions that are compatible with the sample observed, and calculates the likelihood of the data given the current estimate of the haplotype frequencies. For each haplotype the model returns a distribution of frequency estimates from which the mean and its credible interval are derived. For each sample the model returns a distribution, over the possible haplotype compositions with which it is compatible. The model is validated using simulated data sets for which the true haplotype estimates are known. We present results of the application of our model to estimate haplotype frequencies for a set of historic data from Africa in which the prevalence of mutations associated with sulphadoxine-pyrimethamine resistance were obtained.
BibTeX
@inproceedings{taylor2012bayesian,
  author = {Aimee R. Taylor and Jennifer A. Flegg and Philippe J. Guérin and Cally Roper and Chris Holmes},
  title = {{A Bayesian model for estimating with-in host P. falciparum haplotype frequencies}},
  booktitle = {Malaria},
  year = {2012},
  doi = {10.1186/1475-2875-11-s1-p36},
}
Figure