Interpreting Microbial Species-Area Relationships: Effects of Sequence Data Processing Algorithms and Fitting Models

Abstract

In the study of Species-Area Relationships (SARs) in microorganisms, outcome discrepancies primarily stem from divergent high-throughput sequencing data processing algorithms and their combinations with different fitting models. This paper investigates the impacts and underlying causes of using diverse sequence data processing algorithms in microbial SAR studies, as well as compatibility issues that arise between different algorithms and fitting models. The findings indicate that the balancing strategies employed by different algorithms can result in variations in the calculations of alpha and beta diversity, thereby influencing the SARs of microorganisms. Crucially, incompatibilities exist between algorithms and models, with no consistently optimal combination identified. Based on these insights, we recommend prioritizing the use of the DADA2 algorithm in conjunction with a power model, which demonstrates greater compatibility. This study serves as a comprehensive comparison and reference for fundamental methods in microbial SAR research. Future microbial SAR studies should carefully select the most appropriate algorithms and models based on specific research objectives and data structures.

Keywords: high-throughput sequencing; microbial diversity patterns; model fitting; species–area relationship.

Figure 1

SAR curves under different algorithms.

Figure 2

The slopes of the SAR curves under four different algorithms. The slope of the SAR curve, derived from fitting the power model, was analyzed along with p-values obtained through a Kruskal–Wallis test.

Figure 3

Comparison of alpha diversity and beta diversity among four algorithms. (a) displays the overall rarefaction curves for the four algorithms based on species richness; (b) illustrates the overall species abundance rank curve for the four algorithms; (c) presents a beta diversity box plot comparing samples across the four algorithms, with p-values derived from a Kruskal–Wallis test indicating differences; (d) conducts a beta diversity partitioning analysis on eight sample datasets from the four algorithms, where each black dot represents the comparison value between two samples. The positions of the dots are determined by the Richness Difference, Replacement, and Similarity, with the sum of each triplet equaling one. Larger black dots indicate the centroids of the points, representing the average values of Richness Difference, Replacement, and Similarity.

Figure 4

Comparison and verification of algorithm and model fitting results. (a) A scatter plot of AICc values and R² for four algorithms fitted with 20 models, where data with R² values less than zero are considered invalid. The p-value is calculated using the Kruskal–Wallis method to test for differences. (b) A scatter plot of AICc values and R² for the best models across the four algorithms; different colors represent different algorithms, while different shapes denote various models. The p-value is also calculated using the Kruskal–Wallis method to test for differences. (c) The frequency of the best models under the four algorithms, where larger bubbles indicate a greater number of optimal fitting occurrences for each model.