Fig. 1

Conceptual diagram of the process performed by the method proposed in this paper. a The input is the data of two groups X and Y representing the abundance of bacteria. In this figure, we assume that there are 20 kinds of bacteria (indicated by the filled-in circles), and distinguish the classification of each bacteria by phylogenetic annotation using colors (6 colors in this figure). The abundance of each bacteria for each sample is originally given as a count (count_table), which is normalized to a relative abundance (abundance_table) in Step 0. b In Step 1, a bacterial correlation network based on the co-occurrence of bacteria is constructed for each group, by removing noise using existing tools based on the relative abundance of bacteria. In Step 2, in order to eliminate noise due to false correlations, we perform clustering for each of the two groups to unify (contract) the same bacterial groups, and unify the bacterial groups whose phylogenetic annotation classification matches within the same cluster. c Then, for each unified group, the bacteria with the largest average abundance among all the bacteria in that unified group is selected as the representative bacteria of the unified group. The process of contraction to remove false correlations up to this point is the first feature of the proposed method. d Then, in Step 3, the bacteria that significantly increase in group Y compared to group X among the representative bacteria are identified as the core bacteria. In Step 4, a network consisting of core bacteria and their related bacteria is constructed. In the figure, the core bacteria and their related bacteria in the group are shown in yellow background circles, and those in the other group are shown in gray background circles. Their combined network is the final network. e In Step 5, the two networks constructed in Step 4 are compared, and identify bacteria with large QNetDiff scores, which represent the level of rewiring of the links. Then output them together with various statistical values. The process that focuses on rewiring is the second feature of the proposed method