A dataset of rodent-ectoparasite interaction networks and Lotka-Volterra dilution-amplification simulations for 327 sites worldwide, 1934–2020

Of 4,192 records identified, 240 papers reporting usable rodent-ectoparasite interaction data across 327 sites survived title/abstract and full-text screening (documented in the accompanying PRISMA diagram). For each paper, rodent and ectoparasite species names, abundances, and per-individual infestation loads were transcribed into a pre-designed data-extraction template; rodent nomenclature was checked against the IUCN Red List taxonomy and ectoparasite nomenclature against the Global Biodiversity Information Facility.

From these records, one tab-delimited interaction matrix (rodent species x ectoparasite species, cell values = mean ectoparasites per host individual) was built per site. Network-level metrics were then computed for each site's matrix using the networklevel() function (index = "ALLBUTDD") from the R bipartite package. A Bayesian hierarchical model (fit with brms/Stan) was used to relate host and ectoparasite richness to network connectance and to average parasite burden (APB).

Independently, two Lotka-Volterra community models (one with a linear host-parasite functional/numeric response, one with a saturating/ratio-dependent response) were simulated in R using 4th-order Runge-Kutta integration (deSolve) across a grid of host richness (2-14 species) x parasite richness (2-30 species), with randomized interaction networks drawn to match the fitted richness-connectance relationship, to test whether the empirical dilution (host diversity) and amplification (parasite diversity, connectance) effects on APB emerge mechanistically.The empirical data are geographically global: the 327 sites span multiple continents (North and South America, Europe, Africa, and Asia; see the site map in the associated paper), each representing a single field study rather than a regular spatial grid, so spatial resolution is point-based (site-level, with latitude/longitude where available) rather than gridded.

Temporally, the underlying field surveys span 1934-2020 (based on the "Year" field recorded for each source record), reflecting the publication dates and study periods of the 240 source papers rather than a single coordinated sampling campaign; there is no meaningful "temporal resolution" beyond the survey year of each individual study, and no repeated-measures/time-series structure across sites. The simulation output has no spatial or temporal reference at all - it is a set of independent, randomized in-silico community realizations at fixed points in a host-richness x parasite-richness parameter space, not a real-world time series.Several tabular files carry the core content. data/raw_networks/ contains 327 tab-delimited .txt files (about 846 KB total), one per site, each a rodent (rows) x ectoparasite (columns) matrix whose cells give the mean number of that ectoparasite species recovered per individual of that rodent species at that site; row and column headers are binomial species names. data/CPdata_Teng.xlsx (48 KB) is the compiled, one-row-per-site summary table (327 rows x 11 columns): number (site ID), Rodent richness and Ectoparasite richness (species counts), links per species, connectance and weighted connectance (proportion, 0-1, of realized vs. possible host-parasite links), weighted NODF and modularity Q (nestedness/modularity indices), robustness.HL/robustness.LL (host-level/ectoparasite-level network robustness), and log(ectoparasite load) (natural log of APB, exponentiated to APB in the analysis scripts). output/tables/networklevel_all.csv, produced by 01_compute_network_metrics.R, reports 49 bipartite::networklevel metrics (rows) for 326 of the 327 sites (columns, named by source file); it is the fuller, unaggregated counterpart of the metrics summarized in CPdata_Teng.xlsx.

TengSF.xlsx (4.0 MB) is the fuller "Dataset Table" behind the compiled summary, with three sheets: "pre-designed data template" (22,043 data rows, one per rodent-ectoparasite record extracted from the source papers, with columns for source, country, location, year, latitude/longitude, site/network ID, rodent host and its abundance, ectoparasite taxon classification, ectoparasite abundance, and mean load); "Reference List" (240 rows: authors, title, publication year, and abstract for every source paper); and "a pooled rodent-ecto network" (6,010 rows of rodent-ectoparasite species pairs pooled across the entire dataset, i.e. the edge list of the combined global network).

The simulation batches in data/simulations/linear_model/ (5 files, 16 MB total, 56,000 simulated communities each) and data/simulations/saturation_model/ (40 files, 16 MB total, 5,600 simulated communities each) share one schema: Host_sp_num/Pa_sp_num (richness of the simulated community), Avg_Host_competition (mean pairwise host competition coefficient), Connection_rate (realized simulated connectance), Host_suvival_rate/Pa_suvival_rate (fraction of species persisting to equilibrium), Host_biomass/Pa_biomass (summed equilibrium abundance), and Pa_frequency (the simulated APB, i.e. Pa_biomass/Host_biomass, the model analogue of the empirical ectoparasite-load measurements).Two known data issues should be noted before reuse.

First, data/raw_networks/0001.txt contains 11 trailing blank (tab-only) rows and a final stray line containing only the digit "6" after its 4 genuine data rows; this is the only file, of 327, with this problem, and it causes standard whitespace-delimited parsing to fail. 01_compute_network_metrics.R catches and skips unparseable files rather than aborting, so this site's network metrics are currently absent from output/tables/networklevel_all.csv (326 of 327 sites reported) until the source record is corrected or the exclusion is confirmed as permanent.

Second, two auxiliary R scripts in supplementary_exploratory/ (exploratory GLM/model-selection analyses of network metrics against environmental covariates such as temperature, NDVI, and human-modification indices) reference an input file, global.txt, that could not be located and is not included in this deposit; those two scripts cannot currently be re-run, and neither is cited in the manuscript or its Supplementary Information, so they should be treated as unverified side analyses rather than part of the reported results.

Separately, note that the manuscript itself works with several overlapping subsets of the 327 raw sites at different analysis stages (327 raw sites; 299 after excluding the top 1% highest-richness sites as outliers; 239-240 source papers), and users should confirm which subset a given number in the text refers to before comparing it against these files.All tabular files use common, non-proprietary or widely supported formats: plain tab-delimited text (.txt), comma-separated values (.csv), and Excel 2007+ workbooks (.xlsx, openable in Microsoft Excel, LibreOffice Calc, Google Sheets, or programmatically via R's openxlsx/readxl packages or Python's openpyxl).

No rare or proprietary formats are used. The output.zip archive (9.0 MB, 390 files) bundles the data/, output/, supplementary_exploratory/, and deprecated/ folders into a single download and should be unzipped into the same folder as the numbered .R scripts before running any of them. Running the R code requires R (analyses were run under R 4.3.1) plus the CRAN packages listed in 00_install_packages.R (notably bipartite, brms/rstan, deSolve, and ggplot2); fitting the Bayesian model additionally requires a working C++ toolchain for Stan.

Full provenance for every file, including which paper section/figure/table each script and data file corresponds to, is given in README.md.