Daphnia-ms
Mechanistic Models for Panel Data: Analysis of Ecological Experiments with Four Interacting Species
Mechanistic Models for Panel Data
Analysis of Ecological Experiments with Four Interacting Species
Bo Yang, Jesse Wheeler, Meghan A. Duffy, Aaron A. King, Edward L. Ionides
| Preprint: arXiv:2506.04508 | Archive:  |
Overview
This repository contains the code and data to reproduce all analyses in Mechanistic models for panel data: Analysis of ecological experiments with four interacting species. We develop Panel Iterated Filtering (PIF) and Marginalized PIF (MPIF) methods for maximum likelihood inference on PanelPOMP models, and apply them to mesocosm experiments involving two Daphnia species, an algal food resource (Ankistrodesmus falcatus), and a fungal parasite (Metschnikowia bicuspidata).
- Paper & Supplement
- Data
- Model Fitting Code
- Utility Scripts
- Tutorial
- Reproducing the Analysis
- Citation
Paper & Supplementary Material
| Document | Source | Compiled |
|---|---|---|
| Main manuscript | ms.Rnw |
ms.pdf |
| Supplementary material | si.Rnw |
si.pdf |
The manuscript is self-contained and can be generated by knitting ./daphnia-article/ms.Rnw. The supplement in ./daphnia-article/si/si.Rnw is independent of the main manuscript.
Compile the manuscript:
cd ./daphnia-article && Rscript -e "library(knitr); knit2pdf('./ms.Rnw')"
Compile the supplement:
cd ./daphnia-article/si && Rscript -e "library(knitr); knit2pdf('./si.Rnw')"
Note: The majority of expensive computations are done in separate R scripts, but there are a few calculations (including simulation and pomp object creation) that occur within ms.Rnw.
Data
The file Mesocosmdata.xls contains three sheets:
| Sheet | Contents |
|---|---|
| 1 | D. dentifera dynamics with/without parasite |
| 2 | D. lumholtzi dynamics with/without parasite |
| 3 | Mixed D. dentifera and D. lumholtzi dynamics with/without parasite |
Processed data files used by model fitting scripts are in data/. See Section 2 of the manuscript for the full experimental design (6 treatment conditions × 8–10 replicates, 10 time points over 52 days).
Model Fitting Code
All models are fit under the panelPOMP framework using Panel Iterated Filtering. The run_level parameter in each script controls computational intensity:
| Level | Purpose | Approximate Resources |
|---|---|---|
| 1 | Debug | Minutes, any machine |
| 2 | Local computation | ~1 hour on 36 cores, ~50 GB memory |
| 3 | Full estimation | ~40 hours on 36 cores, ~50 GB memory |
Mixed-species treatments (two Daphnia species)
| Model | Treatment | Code | Description |
|---|---|---|---|
| SIRJPF2 | Two species + parasite | Mixed-species/SIRJPF2/ |
Flagship model (8 latent states, 26 parameters) |
| SRJF2 | Two species, no parasite | Mixed-species/SRJF2/ |
Competition through shared food only |
The SIRJPF2 directory contains:
| Subfolder | Contents |
|---|---|
model/ |
Parameter estimation scripts (PIF/MPIF) |
profile/ |
Profile likelihood scripts (MCAP algorithm) |
specific_model/ |
Unit-specific parameter estimation (mif2) |
specific_model_block/ |
Blocked unit-specific estimation (mif2) |
Other model directories follow the same subfolder structure.
Single-species treatments
| Model | Species | Treatment | Code |
|---|---|---|---|
| SIRJPF | D. dentifera | With parasite | Single-species/Dent/SIRJPF/ |
| SIRJPF | D. lumholtzi | With parasite | Single-species/Lum/SIRJPF/ |
| SRJF | D. dentifera | No parasite | Single-species/Dent/SRJF/ |
| SRJF | D. lumholtzi | No parasite | Single-species/Lum/SRJF/ |
Model hierarchy
SIRJPF2 (2 species + Parasite + Food + Juveniles) ← Flagship
├── SIRJPF (1 species + Parasite + Food + Juveniles)
├── SRJF2 (2 species + Food + Juveniles, no parasite)
└── SRJF (1 species + Food + Juveniles, no parasite)
Utility Scripts
The util/ directory contains scripts for generating:
- Profile likelihood plots (MCAP)
- Simulation plots and 95% confidence bands
- Parameter estimation tables
- All figures for the main article and supplementary material
Tutorial
For a step-by-step guide on implementing PanelPOMP models and the methods used in this article, see the companion tutorial:
The tutorial (Supplement 2) covers:
- PanelPOMP model specification
- Panel Iterated Filtering (PIF) for likelihood maximization
- Multi-stage optimization with tempering
- Profile likelihood confidence intervals (MCAP)
- Model diagnostics and simulation studies
Reproducing the Analysis
- Clone this repository:
git clone https://github.com/Megumi-ybb/Daphnia-ms.git cd Daphnia-ms - Install R dependencies:
install.packages(c("pomp", "panelPomp", "tidyverse")) - Fit models (example for the flagship SIRJPF2):
cd Mixed-species/SIRJPF2/model/ # Set run_level in the script (1 = debug, 2 = local, 3 = full) Rscript <script_name>.R - Run profiling:
cd Mixed-species/SIRJPF2/profile/ Rscript <profile_script>.R -
Generate figures and tables using scripts in
util/. - Compile the manuscript and supplement (see commands above).
Citation
If you use this code or data, please cite:
Yang, B., Wheeler, J., Duffy, M. A., King, A. A., & Ionides, E. L. Mechanistic models for panel data: Analysis of ecological experiments with four interacting species. arXiv:2506.04508
BibTeX:
@article{yang2025mechanistic,
title={Mechanistic models for panel data: Analysis of ecological
experiments with four interacting species},
author={Yang, Bo and Wheeler, Jesse and Duffy, Meghan A. and
King, Aaron A. and Ionides, Edward L.},
journal={arXiv preprint arXiv:2506.04508},
year={2025}
}
Software
panelPomp— R package for PanelPOMP models (Bretó et al., 2025)pomp— R package for partially observed Markov processes (King et al., 2016)
License & Acknowledgements
Archived version available on Zenodo: doi:10.5281/zenodo.15595669