JoRussell-IDM
commented
3 years ago Objective: To overview the model structures employed in different GenEpi approaches and the scientific questions best suited to ask with each.
Albert Lee, Jessica Ribado
Glossary:
EMOD: an agent based model of malaria transmission
GenEpi: A model built from tskit to run technical replicates of genetic relatedness under configurable initial diversity on top of externally provided transmission trees with additional functionality to specify a sampling/observational model of the simulated genotypes.
FPG (Full Parasite Genetics): A model built as an extension of EMOD to be able to track parasite genotypes and their recombination within a consistent epidemiological model of malaria transmission with interventions and spatial resolution.
Model approaches.
- EMOD+GenEpi Cotransmission model: A layered model that generates EMOD Transmission Reports (human-vector-human connectivity of infection objects over time) which are subsequently used in GenEpi to simulate vector side recombination and strain level population genetic behavior using the transmission record to include the structure provided by co-transmission events and thus constrain human host connectivity, relatedness, and diversity over time.
Layers:
stochastic seeds of EMOD (technical replicates of transmission network) stochastic replicates of recombination and oocyst numbers in the vector root initialization of neutral markers (barcode SNPs) observation model (samples per human, sites per parasite genome) question for Albert: does genepi require subsampling of simulated genotypes only for post-processing of genetic features? Uses: Most flexible model for testing the independent contribution of model components to genetic outcomes: to test the independent contribution of modeled processes to variance in genetic signals: i.e. How sensitive is the genetic feature Pairwise-IBD to stochastic replicates of the crossovers drawn as part of the recombination model?
Drawbacks: No relevant representation/connectivity of genotype to phenotype (i.e, immunity, drug resistance or HRP2/3 deletions), no gametocyte density dependence for onward transmission, within model responsiveness to genetic signals (interventions in response to surveillance)
- FPG alone model: Everything's baked in (making genetic coinflips alongside epi coinflips), serialization can help in determining random, immunity and specific demography and intervention histories would behave in calibrated sites. This depends on representation of phenotypic markers on our simulated genomes. For var_gene_randomness_types:
ALL_RANDOM: in this case if recombination-specific prng is orthogonal to the remaining prng for DTK-sim (default is whole sim, can be configured for each node which makes it n_cores-invariant)
in absence of drug resistance or any phenotypic tracking this model type (IF FASTER than genepi) could be a useful basis for testing independent contribution of transmission replicates and genetic recombination replicates and observation model
FIXED_MSP or FIXED_NEIGHBORHOOD: genetic decisions inherently impact phenotypic action in the modeled infections and cannot be isolated from epidemiological outcomes, however! If computationally efficient we can run many stochastic replicates of this joint process.
Uses: Intervention impact and
Drawbacks: N
- FPG hybrid: passing of breakpoints to GenEpi. Independence for running efficient replicates for testing Is there anything about the neutral markers signals that is independent of signals we might see in drug resistance.
Uses: (Selective sweep example)
Drawbacks: Ensure that we have consistency in the mathematical representation of crossover and hepatocyte modeled processes to be able to interpret the output of interacting model layers.
Three pipelines GenEpi+DTK
FPG alone: True pairwise IBD - What to show before sharing externally? Mapping sims onto ref epi data from Garki, but do these map onto realistic representations/relationships? Pairwise IBD a function of EIR?
Hybrid: Separate evaluation of neutral sites on a tree of breakpoints. Drug selection sweep for exploring independence of unlinked neutral sites from drug resistance markers on small effective population sizes in the simulation.
Questions of interest to collabs:
Questions:
What is the right threshold for informative contributions to polygenomic infections? (2?) By use case?
What is the right classification for polygenomic infections?
What is sensitivity of genetic metrics to the threshold of observability or classification?
What is a prototypical use case to demonstrate?
Q for Bryan: Is the first step the deconvolution of microhaplotype frequencies across sites? If low diversity between strains, you will underestimate true clonality because of relatedness. Does r refer to relatedness calculated by site or r by strain representation/deconvolution
Q: If you don't know how many unique strains you have (given sequence data from a polygenomic infection), how do you make estimate of COI, without accommodating relatedness in the parasite population you would tend to underestimate COI? Baseline estimate is unique contributors to the diversity at sites. Because of unknown phasing, you will almost certainly be underestimating true COI.
ROLE for MODEL!: Simulate a true COI to help calculate the effective scale factor for converting effective COI to true COI across range of COI (for example effCOI of 2 vs effCOI of 10, also dependent on number of sites and their diversity). Calculation of scale factor may depend on assumptions of transmission intensity and initialization of diversity, structure and linkage of diversity?