Fig. 3

Clustering benchmarks on CAMI datasets. The number of clusters with different recall levels are shown at two precision levels (A CAMI I Medium and High Complexity datasets, B CAMI II Mouse gut datasets, C CAMI II Human Microbiome Project datasets). D Distribution of cluster size at each recall level on CAMI II Human Microbiome Project datasets. All programs were run in multi-sample mode (if have multiple samples) with their default parameters except for MaxBin2 on CAMI I High Complexity dataset. MaxBin2 was run with a single merged sample on this dataset and disabled on the CAMI II Human Microbiome Project datasets due to the excessive runtime with multiple samples. DASTool was run with two combinations: (1) DASTool_MM for MetaDecoder and MetaBAT2; (2) DASTool_CM for CONCOCT and MetaBAT2. MetaDecoder with minimum sequence length setting to 1 Kb (MetaDecoder1000) was also added for benchmarking.