• Software: https://bioconductor.org/packages/FinfoMDS
• File or folder names are italicized. Package or variable names are monospaced. 

File: Data.zip

Description: Raw data used in this study. Includes 4 folders and 4 files (see below).

1. Folder Simulated
   • Contains pairwise distances and ordination results. Includes 6 subfolders and 20 files. See below.
   • Folder F-MDS contains traning log by epoch (folder TrainingLog) and resulting representations Z (folder Results).
     • File names inside the folder are formatted as "sim_rev_{x}-N{n}-{method}-{param}-{type}.csv". Formatting rule is described in table below.
     • "-Z.csv" file is tabulated by each sample and its location in 2D coordinate in each row and column, respectively.
     • "-log.csv" file is tabulated at each row by training epoch. Column obj refers to the value of F-MDS objective. It is the sum of obj_mds the stress term (metric MDS) and obj-confr the confirmatory term weighed by hyperparameter lambda. Columns p-z and p-0 refer to P values computed from data distributions in 2D and original structure, respectively.
   • Folders Isomap, superMDS, t-SNE, UMAP-S, and UMAP-U contain the trained Z from each ordination method.
   • 5 files ending with "-Y.csv" are the response vectors for each training dataset.
     • Similar name rules are applied as "-Z.csv".
     • Note that response vectors are the same for all replicates and their names do not have the replicate number.
   • 15 files ending with "-data.Rds" are the training datasets.
     • Similar name rules are applied as "-Z.csv".
2. Folder Alga
   • Ordination results using microbiome data sampled from laboratory algal mesocosm [1].
   • Microbiome dataset can be found in GitHub package FinfoMDS (https://github.com/soob-kim/FinfoMDS).
   • Contains 8 folders of different ordination methods.
   • File names are formatted similarly to those described for the folder Simulated.
3. Folder HumanGut
   • Contains microbiome data sampled from human gut [2-4] and analyses results.
   • The data is retrieved from a publicly available repository [5], previously merged at the genus level [6].
   • Requires R package phyloseq.
   • File names start with either "cirrhosis-" or "t2d-". It represents the microbiome dataset sampled from patients diagnosed with liver cirrhosis or type 2 diabetes, respectively.
     • "-mds.csv " files are the ordination results using MDS.
     • "-phyloseq.rds" files are the phyloseq object of each community
     • "-Y.csv" files are the data labels ("1" being patients and "2" the control).
4. Folder Ternary
   • Contains 8 files of simulated, ternary data (Fig 7 of main text).
   • File names are formatted similarly to those described for the folder Simulated.
5. File simulated.R
   • Generates binary simulation datasets using SparseDOSSA2 [7] and performs F-MDS.
   • Requires R package vegan, SparseDOSSA2, parallel, and FinfoMDS.
6. File alga.R
   • Performs F-MDS usign algal microbiome dataset.
   • Requires R package parallel, and FinfoMDS.
7. File ternary.R
   • Generates ternary simulation dataset using SparseDOSSA2 [7] and performs F-MDS.
   • Requires R package vegan, SparseDOSSA2, and FinfoMDS.
8. File ordinations.R
   • Produces 2D representation of a dataset using ordination methods.
   • Requires R packages vegan, tsne and uwot.

Table: Formatting rules for data files named "sim_rev_{x}-N{n}-{method}-{param}-{type}.csv".

File: Figure_codes.zip

Description: Codes written in R language for producing main figures and supplementary materials.

File: source.zip

Description: Cloned repositories or packages from other work. Includes one subfolder.

1. Package PopPhy (Reiman et al., 2020) provides a neural network model for classifying algal microbiome. The package was modified in this study by combining with SimCLR framework (Chen et al., 2020), and contains three subfolders.
   • Folder src includes source codes to construct the convolutional neural network as outlined in Reiman et al., 2020. Otherwise specified below, please refer to the authors' documentation.
     • File "config.py" specifies parameter settings for training and evaluating the model.
     • File "train.ipynb" outlines overall process to load / train the data and evaluate the model. Additionally implemented is the contrastive learning as a pretraining stage using library TensorFlow.
     • Subfolder models contains three supporting Python class files to build the models. Additionally included in this work is file "ContrastivePopPhy.py" that performs contrastive learning via SimCLR.
     • Subfolder utils contains seven supporting files that are used to perform the above. Additionally included is "image_augment.py" with functions that adjust data values for self-supervised pretraining.
   • Folder trees includes a reference file of a phylogenetic tree generated by phyloT (see Supplementary Note S3).
   • The algal microbiome dataset can be alternatively found in GitHub package FinfoMDS (https://github.com/soob-kim/FinfoMDS).

References

[1] H. Kim, J. A. Kimbrel, C. A. Vaiana, J. R. Wollard, X. Mayali, and C. R. Buie. Bacterial response to spatial gradients of algal-derived nutrients in a porous microplate. The ISME Journal, 16(4):1036–1045, 2022.

[2] J. Qin, Y. Li, Z. Cai, S. Li, J. Zhu, F. Zhang et al. A metagenome-wide association study of gut microbiota in type 2 diabetes. Nature, 490(7418):55–60, 2012.

[3] N. Qin, F. Yang, A. Li, E. Prifti, Y. Chen, L. Shao et al. Alterations of the human gut microbiome in liver cirrhosis. Nature, 513(7516):59–64, 2014.

[4] F. H. Karlsson, V. Tremaroli, I. Nookaew, G. Bergstrom, C. J. Behre, B. Fagerberg, J. Nielsen, and F. Backhed. Gut metagenome in european women with normal, impaired and diabetic glucose control. Nature, 498(7452):99–103, 2013.

[5] E. Pasolli, D. T. Truong, F. Malik, L. Waldron, and N. Segata. Machine learning meta-analysis of large metagenomic datasets: Tools and biological insights. PLoS Computational Biology, 12(7):e1004977, 2016.

[6] D. Reiman, A. A. Metwally, J. Sun, and Y. Dai. PopPhy-CNN: A phylogenetic tree embedded architecture for convolutional neural networks to predict host phenotype from metagenomic data. IEEE Journal of Biomedical and Health Informatics, 24(10):2993–3001, 2020.

[7] S. Ma, B. Ren, H. Mallick, Y. S. Moon, E. Schwager, S. Maharjan, T. L. Tickle, Y. Lu, R. N. Carmody, E. A. Franzosa, L. Janson, and C. Huttenhower. A statistical model for describing and simulating microbial community profiles. PLoS Computational Biology, 17(9):e1008913, 2021.