Fig. 5. Variation in computing resources affects the irreproducibility of single-gene trees.

Three large representative studies in animals (marine fishes: 1001 genes and 120 taxa from Alfaro et al.⁶⁵), plants (green plants: 410 genes and 1178 taxa from 1KP Initiative⁶⁶), and fungi (budding yeasts: 2408 genes and 343 taxa from Shen et al.¹⁶) were used to examine the impact of threading and processor types on the reproducibility of gene phylogenies. Percentages of the 3819 genes whose phylogenies are irreproducible when a we ran two replicates (Run1 and Run2) on a single node (the two replicates were run one right after the other on the same node) on an increasing number of threads on the CHTC cluster using IQ-TREE (in yellow) and RAxML-NG (in blue), respectively. Since running all 3819 genes on a single laboratory server was computationally intractable, we sampled the first 200 genes from each data set. Percentages of these 600 genes when b we ran two replicates on a laboratory server (Intel Xeon E5–2630 v3 @ 2.40 GHz processor with 16 threads) on an increasing number of threads using IQ-TREE (in yellow) and RAxML-NG (in blue), respectively. Percentages of the 3819 genes whose phylogenies are irreproducible when c we ran two replicates (Run1 and Run2) on two separate nodes (i.e., each analysis was run on a single node, but Run1 was executed on a different node than Run2) on an increasing number of threads on the CHTC cluster using IQ-TREE (in yellow) and RAxML-NG (in blue), respectively. The irreproducibility of each gene was determined by comparing the topologies of single-gene trees generated by two replicates (Run1 and Run2). These results suggest that multithreading, coupled with the use of different nodes, is a major contributing factor to irreproducibility in IQ-TREE and that the use of different nodes, but not multithreading, is a major contributing factor to irreproducibility in RAxML-NG. Command lines and job submission files are given in Supplementary Note 1. All gene trees, log files, and statistics of the results, are available on the figshare repository: 10.6084/m9.figshare.11917770.