Initial data
The input is a semicolon-delimited CSV file containing a species-by-sample matrix. The first row contains sample names, and the first column contains species names. The file may contain quantitative or presence-absence data; all values are converted to presence or absence before analysis.
Analysis
Associations are identified using hypergeometric confidence intervals, binomial confidence intervals, or beta credible intervals.
In the results table, SP1 and SP2 identify the two species; N1 and N2 are their occurrences; and N1N2 is their observed joint occurrence. p.lt and p.gt are the lower- and upper-tail probabilities, or the corresponding posterior tail probabilities for the beta method. A small p.lt supports a negative association, whereas a small p.gt supports a positive association.
Graph
The graph contains significantly associated species pairs. Node size is logarithmically proportional to species occurrence. Edge thickness increases as the relevant tail probability decreases. The graph can include positive associations, negative associations, or both.
The Optimal Community algorithm uses the GLPK package to find a partition with maximum modularity. Computation time may increase rapidly with graph size.
When negative associations are included, signed Spin-glass is the only available clustering method. It treats positive associations as attraction and negative associations as repulsion.
Cluster statistics
Connectivity is the proportion of all possible within-cluster edges that are present. Mean connection strength is the mean of 1 - tail probability / significance level across the observed within-cluster edges. Total connectivity strength is the sum of these strengths divided by the number of all possible within-cluster edges; absent edges therefore contribute zero.
About
Method presentation (PPTX in Russian).
Method description by V. K. Shitikov (PDF in Russian, with R code examples).
Methodology for the identification and analysis of species associations in biological communities using statistical distributions (English preprint at bioRxiv).
The test dataset was taken from Ibragimova et al., 2025, Part I. DOI: 10.15298/arthsel.34.2.05.
Contact the author: Dmitriy Seleznev.
Hosted by the Papanin Institute for Biology of Inland Waters, Russian Academy of Sciences.