Read on Twitter
Mosquito populations are highly dynamic over space and time; they shape the temporal profile of malaria risk which impacts when and how we deploy control interventions.
Understanding them is therefore pretty important. Our work now up on medRxiv: https://www.medrxiv.org/content/10.1101/2021.01.09.21249456v1 /1 https://twitter.com/DrSamirBhatt/status/1349111371204538370
Understanding them is therefore pretty important. Our work now up on medRxiv: https://www.medrxiv.org/content/10.1101/2021.01.09.21249456v1 /1 https://twitter.com/DrSamirBhatt/status/1349111371204538370
Lots of amazing work has been done mapping the spatial presence/absence of mosquito vectors of malaria (e.g. https://tinyurl.com/yyanhtmu ), but comparatively little work has focussed on systematically understanding their temporal dynamics. /2
To try and better understand these dynamics and the factors driving them, we carried out a systematic review of published mosquito catches spanning at least 1 year across India. Results from 7 major malaria vectors, 40 years and 117 unique locations. /3
What really stood out was the variation- An.dirus populations tended to peak during the wet season (roughly June to September) whilst for An.fluviatilis, this was typically in the dry season. For An.culicifacies, we saw everything from perennial dynamics through to sharp peaks /4
To each of these time-series, we fitted Negative Binomial GPs with periodic kernels in
@mcmc_stan
, and inspired by work from
@bendfulcher
( https://royalsocietypublishing.org/doi/10.1098/rsif.2013.0048), used a suite of statistical methods to characterise temporal properties and identify latent structure. /5
@mcmc_stan
, and inspired by work from
@bendfulcher
( https://royalsocietypublishing.org/doi/10.1098/rsif.2013.0048), used a suite of statistical methods to characterise temporal properties and identify latent structure. /5
These properties range from simple metrics like “Number of Peaks” through to KL Divergence from a uniform and the period of our GP kernel. We do some Von Mises fitting along the way too - shout out to Michele Nguyen and MAP for inspiring much of this: /6 https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-019-1486-3
This allowed us to group together time-series with similar properties – clustering consistently identified 4 key groups– each with distinct temporal properties as measured by timing and extent of seasonality. Wet season, dry season, bimodal, perennial; we’ve got the lot /7
(this is literally the only time ever that a PCA + k means actually worked for me to produce nice, distinct groups, so that's nice too) /8
So why all this variation? We’re still figuring that bit out, but we’ve got some clues – all our studies were geolocated and so we were able to regress a bunch of environmental covariates (as well as species as a variable) on these cluster labels. /9
Key Takehome #1: Some species show very strong associations with a particular temporal modality. Despite some noise, Anopheles fluviatilis was consistently associated with dry season dynamics irrespective of the location (and broader environment) the population resided in. /10
(Here's all our species-complexes hierarchically clustered according to their patterns of association with particular temporal patterns) /11
Key Takehome #2: The ecology of a setting influences dynamics. E.g. Perennial abundance was consistently associated with proximity to large, static bodies of water, whilst wet season peaking dynamics were associated with temperature seasonality and total rainfall. /12
Key Takehome #3: These associations were frequently exclusive – variables (e.g. total rainfall) tended to strongly and specifically associate with a particular temporal pattern (Cluster 1), highlighting that unique sets of ecological factors drive different temporal profiles. /13
We’re still a long way off a proper finished product, but a better understanding of these relationships allows us to begin to predict spatial variation in the temporal modalities mosquito populations display, as shown through these map. /14
(Doing this also fulfilled a personal PhD dream of mine to make pretty maps that hopefully, vaguely mean something, so that was nice too!) /15
Still a lot to learn, but we’re hoping this is the beginnings of a framework enabling systematic characterisation of time-series data relevant to malaria control like this. Our hope is to test this on vector data from a wider range of locations, as well as… /16
…begin integrating malaria case data alongside the entomological data. We’d love to hear from you if you have that data and would be interested in working together and collaborating!!! /17
