Thousands of taxa, a handful of states: Arctic protist communities follow a few environmental drivers.
Draft submission package with source code and data for this publication.
Authors: Alexey Ryabov* , Rahel Vortmeyer-Kley, Bernd Blasius, Thilo Gross, Rebecca McPherson, Sinhué Torres-Valdés, Katja Metfies*.
Please contact Alexey Ryabov (ICBM, Uni Oldenburg) or Katja Metfies (AWI) if you want to use the data.
The MATLAB source covers Fig. 2–3 and 5–7 and Fig. S1–S11. Fig. 1 and 4 are outside this package.
source/: figure scripts, project configuration, and required local helpers.data/: the input files read by the originaldt_loadData, plus retained station coordinates.source/dt_loadData.mretains its filtering and normalization. The selected settings areDataID = 'HG_20092023'andJobID = 'ASVNoMetazoaJ'.results/: generated output location. It is excluded by.gitignore; a fresh clone starts with an emptyresults/.
The four retained data files are HGSamples.mat, Samples_imp.xlsx,
Phyla_roles_and_sizes.xlsx, and Stations_coorsd.xlsx. The first three
are read by dt_loadData; station coordinates are retained but are not
used by the current scripts.
HGSamples.mat and Samples_imp.xlsx are prepared upstream data, not raw
sequencing reads or unimputed environmental observations. Their upstream
preparation is outside the scope of this package.
- MATLAB R2025b was used to check this package. A static dependency scan also identified Statistics and Machine Learning Toolbox, Bioinformatics Toolbox, Financial Toolbox, and Parallel Computing Toolbox. The exact product set needed for each figure has not yet been verified by a clean run.
- Diffusion map is an external dependency. Install it separately; its functions are not included in this repository.
- Violin Plots for Matlab
must be installed separately for
Fram_0273_Violins_CommComposition.m. The installation must provide bothviolinplot.mandViolin.m. source/lib/contains the other functions found by the dependency scan.bcdistcorr.mis distributed with its original license insource/lib/LICENSE-bcdistcorr.txt.
Run scripts with source/ as the MATLAB current folder. Initialization adds
source/lib/ to the MATLAB path. DT_Params.m resolves the input files
relative to this package's data/ folder. Running proj_Initialize.m
performs the original filtering and writes tSamplesNorm.xlsx and
tStPredictors.xlsx under results/HG_20092023/ASVNoMetazoaJ/.
| Figure | Script(s) |
|---|---|
| 2 | Fram_0260_DMClusterPlot.m |
| 3a–c | Fram_0310_SVM_Type_Init_LonMonth.m, Fram_0322_SVM_Type_Train_LonMonth.m, Fram_0330_SVM_Type_PD_2D_LonMonth.m |
| 3d–f | Fram_0350_ML_Init_TempDepDoy.m, Fram_0352_ML_Train_TempDepDoy.m, Fram_0354_ML_Plot_TempDepDoy.m |
| 3g, 5 | Fram_0273_Violins_CommComposition.m |
| 6 | Fram_0296_CoreContribFinalClean.m |
| 7 | Fram_0347_Chl_LongTempIceSeasonUpper30m.m |
| S1, S2 | Fram_0100_DM.m |
| S3 | Fram_0105_DM_Cluster_Test.m |
| S4a,b | Fram_0310_SVM_Type_Init_LonMonth.m, Fram_0322_SVM_Type_Train_LonMonth.m (PredictorSetID = 'LonMonthDepth') |
| S4c,d | Fram_0350_ML_Init_TempDepDoy.m, Fram_0352_ML_Train_TempDepDoy.m |
| S5 | Fram_0310_SVM_Type_Init_5drivers.m, Fram_0322_SVM_Type_Train_5drivers.m, Fram_0332_SVM_Type_PD_1D_5drivers.m (PredictorSetID = 'TempPARDepthNitrateIce') |
| S6 | Fram_006_Env.m |
| S7 | Fram_0273_Violins_CommComposition.m |
| S8 | Fram_0250_Cluster.m |
| S9 | Fram_0275_CommCoreSpec.m |
| S10, S11 | Fram_0296_CoreContribFinalClean.m |
The diffusion map from Fram_0100_DM.m is an input to clustering; the cluster output is an
input to several later scripts. Fram_0275_CommCoreSpec.m creates the core
species data used by Fram_0296_CoreContribFinalClean.m.
For a fresh clone, run Fram_0100_DM.m before Fram_0250_Cluster.m.
The latter creates the cluster labels needed by the later figure and model
scripts. Run Fram_0275_CommCoreSpec.m before
Fram_0296_CoreContribFinalClean.m.
Train each SVM before running its partial dependence plot script. Partial
dependence values are not included in the repository: they must be calculated
by running the plot scripts. All three partial dependence scripts set
calcPlot = true, so a fresh clone calculates the values rather than
loading absent cached files. The calculated values are saved under
results/ for later reuse.
The LonMonth and 5drivers chains use separate initialization and training scripts; each plot script loads the model trained with its own predictor set.