CHAMP delivers accurate taxonomic profiles of the prokaryotes, eukaryotes, and bacteriophages in the human microbiome

Abstract

Introduction: Accurate taxonomic profiling of the human microbiome composition is crucial for linking microbial species to health outcomes. Therefore, we created the Clinical Microbiomics Human Microbiome Profiler (CHAMP), a comprehensive tool designed for the profiling of prokaryotes, eukaryotes, and viruses across all body sites.

Methods: CHAMP uses a reference database derived from 30,382 human microbiome samples, covering 6,567 prokaryotic and 244 eukaryotic species, as well as 64,003 viruses. We benchmarked CHAMP against established profiling tools (MetaPhlAn 4, Bracken 2, mOTUs 3, and Phanta) using a diverse set of in silico metagenomes and DNA mock communities.

Results: CHAMP demonstrated unparalleled species recall, F1 score, and significantly reduced false positives compared to all other tools benchmarked. The false positive relative abundance (FPRA) for CHAMP was, on average, 50-fold lower than the second-best performing profiler. CHAMP also proved to be more robust than other tools at low sequencing depths, highlighting its application for low biomass samples.

Discussion: Taken together, this establishes CHAMP as a best-in-class human microbiome profiler of prokaryotes, eukaryotes, and viruses in diverse and complex communities across low and high biomass samples. CHAMP profiling is offered as a service by Clinical Microbiomics A/S and is available for a fee at https://cosmosidhub.com.

Keywords: MAG; bacteriophages; benchmarking; human microbiome; metagenomics; taxonomic profiling.

Figure 1

Evaluating the taxonomic profilers CHAMP, MetaPhlAn 4, mOTUs 3, and Bracken 2 on key benchmarking metrics: precision, recall, F1 score, similarity, and FPRA across five human body communities. A total of 50 metagenomes (10 per body site) each with 100 prokaryotic species were simulated using CAMISIM.

Figure 2

Evaluating performance across the sampled sequencing depths: 20, 5, 2, 1, 0.5, 0.25 and 0.1 million reads for the four taxonomic profilers: CHAMP, MetaPhlAn 4, mOTUs 3, and Bracken 2. The following benchmarking metrics were compared: precision, recall, F1 score, similarity, and FPRA. Each of 50 (10 per body site) metagenomes comprising 100 species were randomly sampled at seven sequencing depths resulting in 350 metagenomic samples.

Figure 3

Comparison of eukaryotic taxonomic profiling by CHAMP and MetaPhlAn 4 using in silico metagenomes. 30 metagenomes were simulated encompassing 113 eukaryotic species found both in CHAMP and MetaPhlAn 4 databases. Benchmarking metrics included: precision, recall, F1 score, similarity, and FPRA.

Figure 4

(A,B) Benchmark of phage profiling on in silico dataset with 10 simulated metagenomes containing 95% prokaryotic and 5% phage reads across the sequencing depths: 10, 20, 25, and 50 million reads. The performance of phage profiling with CHAMP and Phanta was assessed by F1 score, precision, recall, similarity, and FPRA. (C) Precision and recall of CHAMP after rarefying the 10 in silico communities with 50 million to 25, 20, 10, 5, 2, 1, 0.5, 0.25, and 0.1 million reads.