Abstract
In microbiome analysis, functional profiling is based on assigning reads or contigs to terms or nodes in a functional classification system. There are a number of large, general-purpose functional classifications that are in use, such as eggNOG, KEGG, InterPro and SEED. Smaller, special-purpose classifications include CARD, EC, MetaCyc and VFDB. Here, we compare the different classifications in terms of their overlap, redundancy, structure and assignment rates. We also provide mappings between main concepts in different classifications. For the large classifications, we find that eggNOG performs the best with respect to sequence redundancy and structure, SEED has the cleanest hierarchy, whereas KEGG and InterPro:BP might be more informative for medical applications. We illustrate the practical assignment rates for different classifications using a number of metagenomic samples.
INTRODUCTION
In microbiome analysis, functional profiling aims at answering the question of ‘what can microorganisms in this sample do?’. One straight-forward way to address this is to align microbiome shotgun sequencing reads (or their assemblies) against functional reference sequences and then to bin them by function (1,2).
There exist a number of large functional classification systems that can be used for general-purpose analysis, such as eggNOG (3), KEGG (4), InterPRO families (5) or SEED (6). Specific questions are sometimes addressed using more special-purpose databases. For example, Enzyme (EC numbers) (7) and MetaCyc (8) can be used to analyze metabolism, whereas CARD (9), MEGARes (10) and ARGminer (11) address antimicrobial resistance, and VFDB (12) focuses on virulence factors.
How do these approaches differ from each other? To compare different functional classification systems, we first looked at their sequence content, that is, the sequences contained in the databases associated with each hierarchy. For each sequence in one database we searched for its best match in another.
Matching can be done by sequence similarity or semantically. Semantic comparison matches proteins by their annotated function, based on assigned GO terms, say, and does not always correlate strongly with sequence similarity. So, potentially, semantic search allows one to find matches that are not detectable by homology search (13). However, since metagenomic data analysis is usually based on homology search methods, such as BLAST (14) or DIAMOND (15); here, we will match sequences based on similarity only. To avoid confusion we did not use any mapping between sequences or terms in databases that are provided by their curators. Instead we performed DIAMOND alignment of sequences present in one database against the other.
We compared four of the most widely used general-purposes databases so as to determine how they cover and complement one another, namely InterPro families (mapped to the GO:Biological Processes part of the metagenomic GO-slim (16), referred to here as InterPro:BP), eggNOG, KEGG pathways and SEED. We explored how well concepts in one system map onto concepts in the other systems. In addition, we compared the assignment rates for the four databases on five different short read datasets sampled from different environments, as calculated using MEGAN6 (17).
Additionally, we investigated how the six mentioned special-purpose classification systems fit into the four presented general-purpose ones.
We found that eggNOG performs the best with respect to sequence redundancy and structure, whereas, KEGG, SEED and InterPro:BP have much higher sequence redundancy. Moreover, the GO:Biological Processes classification, as provided by in the metagenomic GO slim, has a structure that is too general for good overviews.
General purpose functional classifications
Three of the above mentioned general-purpose classifications (eggNOG, KEGG and SEED) are hierarchically structured whereas some, but not all, of the InterPro families are connected among themselves. In order to get a complete hierarchy, they are connected with GO (18). In each hierarchy root nodes correspond to the database itself, whereas leaves represent the most specific terms and have reference sequences associated with them. This makes it easy to provide overviews at different levels of detail. The most specific terms in eggNOG and KEGG are othologous groups, in InterPro:BP it is protein families and subsystems in SEED. Table 1 summarizes main statistics of the four hierarchies.
Table 1.
Basic statistics. For each of the classifications eggNOG, KEGG, InterPro:BP and SEED, we report the median depth (that is, the median number of nodes on the path from the root to leaves), the number of internal nodes and leaves, and the number of reference sequences
| eggNOG | KEGG | InterPro:BP | SEED | |
|---|---|---|---|---|
| Median depth | 4 | 5 | 3 | 5 |
| Internal nodes | 28 | 591 | 1528 | 154 |
| Leaves | 30 955 | 55 124 | 9581 | 823 |
| Sequences | 7.5M | 13.2M | 14.8M | 47.7M |
Protein families include all evolutionary related proteins, namely both othologs and paralogs. While othologs arise due to speciation events and contain proteins that share the same function, paralogs emerge by gene duplication, with a possible subsequent change of function (19). Therefore, orthologous groups are by nature more conserved than protein families. Subsystems, as used by SEED, are even more general than protein families and are more similar to pathways. Each subsystem contains a set of proteins that realize a specific biological process or a structural complex (20). This is reflected in the statistics reported in Table 1. EggNOG and KEGG have fewer sequences, but significantly more leaves than InterPro:BP and SEED, whereas SEED has fewest leaves (subsystems), but most sequences.
All four databases are based on data from various sources as well as on each other. EggNOG which is an acronym for ‘Evolutionary genealogy of genes: Non-supervised Orthologous Groups’ is compiled from genomes that undergo checks for quality and completeness. In addition it contains some manually curated orthologous groups (3). Each orthologous group is assigned a function using a sophisticated pipeline and many different sources including GO (18), KEGG (4) and placed in a functional category as introduced in COG (21), KOG (22) and arCOG (23).
The Kyoto Encyclopedia of Genes and Genomes (KEGG) is a collection of 18 specific databases that are all manually curated, except for one computationally generated database (SSDB) (24). KEGG contains information from fully sequenced genomes as well as individual proteins with experimentally characterized function (4). Here, we considered all KOs from the KEGG othology database except for those of the BRITE hierarchies.
InterPRO families is one of the largest databases for automatic sequence annotation. InterPRO contains 27.5M sequences from 14 other databases (5): CATH-Gene3D (25), CDD (26), HAMAP (27), PANTHER (28), Pfam (29), PIRSF (30), PRINTS (31), ProDom (32), PROSITE Patterns (33), PROSITE Profiles (33), SMART (34), SFLD (35), SUPERFAMILY (36) and TIGRFAMs (37). Data from these databases are manually integrated into InterPRO and a GO number is assigned. We analyzed InterPRO families as mapped to GO slim terms used for metagenomic analyses in MEGAN (17). As explained below in the methods section, of the three ontologies available in GO, we only consider Biological Process.
The SEED contains protein families that are compiled from various resources and contains genes from published genomes only. It uses public genomes annotated by the RAST (Rapid Annotation using Subsystem Technology) pipeline (38), expert user annotations, metabolic modeling data (39,40), expression data and literature references verifying annotations (41). SEED is widely used both for microbial genome annotation in the RAST pipeline and for metagenomic analysis in MG-RAST (2). The hierarchy of functional classification terms behind the SEED is based on subsystems as introduced in (20). The most recent version of it is included in PATRIC (Pathosystems Resource Integration Center) (42) as Subsystems.
Special-purpose functional classifications
MetaCyc and Enzyme (EC numbers) are two widely used databases (and implicit classifications) that are dedicated to metabolism. MetaCyc is a well curated and evidence-based database of metabolic pathways and enzymes. It contains proteins from experimentally determined and published pathways. Because the curators apply strict criteria, it only contains 12K sequences. In comparison, the Enzyme database contains over 230K proteins. EC numbers are recommended by the Nomenclature Committee of IUBMB (International Union of Biochemistry and Molecular Biology) (43) and the database contains both confirmed and preliminary EC numbers.
Antibiotic resistance is an important medical category. There are a number of databases that address this. Identifying ARGs (antibiotic resistance genes) in a microbiome setting requires a high quality database of reference sequences and a dedicated analysis pipeline (44). The prediction of antibiotic resistance is more challenging than the general-purpose functional profiling discussed above. Antibiotic resistance depends on a number of different features, such as the presence, or absence, of a certain protein variant, or DNA sequence, or SNP, or on the order of genes along the genome, say. CARD (Comprehensive Antibiotic Resistance Database) is the smallest ARG database with 2.6K sequences. It is highly curated and contains only antimicrobial resistance determinants with clear experimental evidence (9). MEGARes contains almost 8K sequences and is also manually curated. It is specifically intended for high-throughput data, which makes it especially applicable to metagenomics studies. The MEGARes database is compiled from non-redundant proteins obtained from CARD (9) and other sources (10). ARGminer contains almost 15K protein sequences (11). It is a community curated resource that is dependent on several other resources, including both CARD (9) and MEGARes (10).
A further topic that is also relevant to medical research is pathogen identification. VFDB (Virulence Factor Database) is a database dedicated to this question. It is well curated and contains data from 32 medically important pathogens. It is available in two versions. The ‘core’ dataset contains only experimentally validated virulence factors, whereas the ‘full’ dataset contains all genes related to known and predicted virulence factors (12).
MATERIALS AND METHODS
Data download
Sequence data for functional classification systems was obtained from the following databases: the NCBI non-redundant protein database (NCBI-nr) downloaded January 2021 from ftp://ftp.ncbi.nih.gov/blast/db/FASTA/nr.gz, InterPRO downloaded January 2021 from ftp://ftp.ebi.ac.uk/pub/databases/interpro, eggNOG version 4.5 downloaded January 2021 from http://eggnogdb.embl.de, KEGG orthology downloaded January 2021 from http://rest.kegg.jp, SEED as extracted from PATRIC 3.6.9, Enzyme downloaded July 2020, MetaCyc 24.0, CARD downloaded August 2020, VFDB downloaded August 2020, MegaRES v2.0, ARGMiner v1.1.1.A, ACLAME v0.4. The metagenomic GO slim was downloaded January 2021 from http://geneontology.org/ontology/subsets.
The three GO domains and InterPro
The GO gene ontology consists of three separate domains: biological process, cellular component and molecular function (18). All three are also represented in the metagenomic GO-slim used here (16). MEGAN uses a hierarchy based on the metagenomic GO-slim to classify InterPro families. To determine how redundant the representation of InterPro families by the three different domains is, we determined the percentage of proteins in InterPro families that they share, as shown in Figure 1.
Figure 1.
Venn diagram shown the percentage of InterPro family sequences assigned to the three GO domains (biological process, cellular component and molecular function).
The two domains biological process and molecular function contain 95% of all sequences present in InterPro families. Molecular function contains 17M sequences whereas biological process over 15M. Because the GO:Biological Process ontology is conceptually the most similar to other functional classification systems discussed here, we only consider this GO ontology.
Throughout this paper, we use the term InterPro:BP to refer to the functional classification that uses InterPro families as leaves and terms of the metagenomic GO-slim as internal nodes, as introduced in (17), restricted to the GO:Biological Process domain.
Database redundancy
To determine redundancy, databases discussed in the paper were clustered using CD-HIT (45) version 4.8.1 at a sequence identity threshold of both 0.9 and 0.7, with a length difference cutoff of 0.9 and minimal alignment coverage of 0.9. To verify the obtained number of clusters, USEARCH (46) v11.0.667 was additionally applied to the small databases. On average USEARCH found approximately 0.4% more clusters than CD-HIT (see Supplementary Table S1).
Comparison of functional classification systems by read mapping
Five examples of short read microbiome shotgun-sequencing datasets were downloaded from the NCBI Sequence Read Archive (SRA) (47), see Table 2. All five datasets were preprocessed with fastp (v0.21.0) (48), with the base quality threshold set to 20 and all other parameters using default values.
Table 2.
Information on short read data. For each dataset, referred to by its environmental source, we report the Biosample ID, number of reads after quality filtering and average read length
| Dataset | Biosample | No. reads | Length |
|---|---|---|---|
| Gut | SAMEA4731558 | 16.3M | 150 |
| Skin | SAMEA4611035 | 22.8M | 100 |
| Water | SAMEA2421979 | 0.6M | 511 |
| Landfill1 | SAMN14103086 | 15.7M | 150 |
| Landfill2 | SAMN13336098 | 25.3M | 151 |
To explore the assignment rate of each functional classification system in practical metagenomic analysis, we analyzed the short read datasets using both the general-purpose ‘DIAMOND against NCBI-nr + MEGAN’ pipeline and also a dedicated ‘DIAMOND against a specific database + MEGAN’ pipeline, for each specific database, using DIAMOND (15) (v2.0.7) in BLASTX mode and MEGAN6 (v6.21.1) (17) (default parameters).
To compare read mapping to different functional classification systems we used DIAMOND (query cover set to 90%, otherwise default parameters) retaining only the best match. For each pair of functional classifications, we determined the proportion of reads that could be mapped in both classifications, or only one of them.
Comparison of functional classification systems by sequence content
To compare functional classification systems as a whole, we mapped all sequences contained in one database on to all others. For consistency, we ignored any mappings provided by the authors of the classification systems, and instead compiled DIAMOND indices for each functional classification system, and then used DIAMOND in BLASTP mode to match sequences. For each sequence only one best hit with at least 90% of query coverage was considered (with default values for all other DIAMOND parameters).
Functional classification systems were compared in pairs. First we determined which proportion of sequences from one database could be mapped on to another. To compare hierarchies of functional classifications, we considered selected nodes or terms, together with all sequences assigned to them and all their descendants. For the different classifications, we decided which hierarchy level to consider based on how general or specific it’s nodes are. We aimed to select nodes that were general enough to provide an overview of the data while remaining functionally specific. Thus, for eggNOG and KEGG we considered the second level below the root, for GO:Biological Process the first level and for SEED a mixture of levels that best matched the level of detail of the other three functional classifications. In eggNOG the choice corresponds to functional categories, in KEGG to smallest groups of pathways, in InterPro:BP to GO terms as available in the metagenomic GO slim whereas in SEED it was a set of superclasses, classes and subclasses as available in the subsystems of PATRIC.
For each pair of nodes from two different classifications, we determined the proportion of sequences assigned to the one node (and its descendants) that had matches to sequences assigned to the other node. The proportion of sequences that could not be assigned to any of the nodes was also determined.
Note that matching between functional classification systems is usually not symmetric because the number of sequences assigned to the compared nodes may be very different, and because sequence alignment is not symmetric.
RESULTS
Size and redundancy of functional databases
To determine the redundancy of each of the four main functional classification systems available in MEGAN (InterPro:BP, eggNOG, KEGG and SEED), we clustered all sequences using a sequence identity threshold of both 90% and 70%. Figure 2 A shows the difference between the number of sequences and the number of clusters. The lower the identity threshold is, the smaller the difference in size among the four large functional classification systems is. For example, SEED has more than six times as many sequences as eggNOG, but only three times as many clusters with 90% and twice as many with 70% sequence identity.
Figure 2.
(A) For four large functional classification systems, we report the number of reference sequences, and the number of clusters obtained using CD-Hit with a sequence identity cutoff of 90% or 70%, respectively. (B) For various functional classification systems, we report the ratio of the number of clusters to reference sequences, for clusters obtained using CD-Hit with a sequence identity cutoff of 90% or 70%, respectively.
In Figure 2B, we present the clusters-to-sequences ratio in all functional classification systems discussed here (large and small). The number of clusters obtained (for both 90% and 70% identity) indicates that eggNOG has more diversity than KEGG even though it contains less sequences in total. SEED and InterPro:BP both have more clusters than eggNOG; however, SEED has a very low sequence to cluster ratio, indicating a high percentage of sequence similarity and low diversity for the total number of sequences. For exact numbers, please see Supplementary Table S2.
Both MetaCyc and the core of VFDB are small and strictly curated databases, containing only proteins that have been experimentally validated In consequence, they have a high cluster-to-sequence ratio of over 0.96, for 90% identity, and over 0.91, for 70%, respectively. Databases for antimicrobial resistance have the highest redundancy. This is because all variants are of high importance and only hits with very high sequence similarity are of interest.
Comparison of functional classifications
Table 3 shows the percentage of sequences in each of the four large functional classification systems that can be mapped to each of the others. From this general comparison it appears that InterPro:BP contains most of the information that is present in eggNOG, KEGG pathways and SEED. However, these numbers say nothing about structural similarities among the four. The heatmaps presented in Figure 3 and Supplementary Tables S3–S8 provide a more detailed comparison.
Table 3.
For each functional classification system listed on the left, we report how well it maps to the classification systems listed along the top, in other words, the percentage of sequences that are covered by sequences in the system listed on the top
| InterPro:BP | eggNOG | KEGG | SEED | |
|---|---|---|---|---|
| InterPro:BP | — | 68.1 | 59.2 | 55.9 |
| eggNOG | 97.7 | — | 65.3 | 70.6 |
| KEGG | 97.2 | 75.4 | — | 64.2 |
| SEED | 98.7 | 97.3 | 82.7 | — |
Figure 3.
Heatmaps showing how nodes from functional classifications on the left are mapped to the nodes of functional classifications along the top. White indicates 0% sequences, dark red—100%, on a linear scale. The last column in each heatmap shows the proportion of the sequences that cannot be matched to any node. Boxes highlight the nodes that correspond to metabolism (or metabolic process in InterPro:BP). More details can be found in the Supplementary Tables S3–S8.
Note that a high percentage of SEED is covered by both eggNOG and InterPro:BP but not the other way around. This is also the case for KEGG, although not as much of it is covered by eggNOG as is by InterPro:BP. This suggests that eggNOG and InterPro:BP are more general than SEED and KEGG, and so we first discuss similarities and differences between eggNOG and InterPro:BP
Comparison of eggNOG and InterPro:BP
Even though eggNOG maps much better to InterPro:BP, than vice versa, the structure of InterPro:BP appears to be uninformative. Almost all of the nodes map best to GO:0008152 Metabolic Process (approximately 50% sequences per node) and GO:0006810 transport (approximately 20% of sequences per node) (see Supplementary Table S3A). Only two exceptions are present in eggNOG, that is Extracellular Structures and Cytoskeleton. Bot of these nodes have a large proportion (52% and 44%, respectively) of sequences that do not have matches in InterPro:BP. The node GO:0008152 Metabolic Process contains almost 100 other GO terms and InterPro families that are much more specific than eggNOG terms and thus are not suited for a comparison. The structure of InterPro:BP lacks intermediate levels and goes directly from very general to very specific terms, making InterPro:BP less useful than eggNOG for functional overviews.
We observed the same trend for mappings of KEGG and SEED to InterPro:BP (see Supplementary Tables S4 and S5). Therefore we used eggNOG as reference database for exploring KEGG and SEED and will not discuss their similarity to InterPro:BP in detail.
The mapping of InterPro:BP to eggNOG is also very poor (Supplementary Table S3B). Most of the terms are largely unassigned whereas matches that can be found are very weak (less than 30% of sequences), as shown in the Table 4. InterPro:BP terms that have >65% of unassigned sequences are GO:0009405 Pathogenesis (68% unassigned), GO:0016032 Viral Process (74%) and GO:0046718 Viral Entry into Host Cell (73%). This indicates that eggNOG may lack information present in InterPro:BP that is important for medical applications.
Table 4.
Matching of InterPro:BP terms to eggNOG terms. For each pair, we report the percentage of reference sequences for a InterPro:BP node that are highly similar to some reference sequence for the corresponding eggNOG node. Matches are shown only for those nodes in InterPro:BP that had a best match in eggNOG covering at least 15% of sequences contained in it. The complete list is provided in Supplementary Table S3B
| InterPro:BP | eggNOG | % |
|---|---|---|
| GO:0006259 DNA metabolic process | Information storage and processing → [L] Replication, recombination and repair | 27 |
| GO:0006457 protein folding | Cellular processes and signaling → [O] Post-translational modification, protein turnover, chaperones | 27 |
| GO:0006808 regulation of nitrogen utilization | Metabolism → [E] Amino acid transport and metabolism | 28 |
| GO:0006950 response to stress | Information storage and processing → [L] Replication, recombination and repair | 21 |
| GO:0008152 metabolic process | Metabolism → [C] Energy production and conversion | 16 |
| GO:0016070 RNA metabolic process | Information storage and processing → [J] Translation, ribosomal structure and biogenesis | 17 |
| GO:0016226 iron-sulfur cluster assembly | Cellular processes and signaling → [O] Post-translational modification, protein turnover, chaperones | 23 |
| GO:0019222 regulation of metabolic process | Information storage and processing → [K] Transcription | 15 |
| GO:0045454 cell redox homeostasis | Cellular processes and signaling → [O] Post-translational modification, protein turnover, chaperones | 30 |
| GO:0071973 bacterial-type flagellum-dependent cell motility | Cellular processes and signaling → [N] Cell motility | 33 |
Comparison of KEGG and eggNOG
About half of the considered nodes in KEGG have good matches in eggNOG. As shown in Table 5, most of the terms under Metabolism in KEGG map well to Metabolism terms in eggNOG, whereas genetic information processing in KEGG maps to information storage and processing in eggNOG. As can be seen in the heatmap in Figure 3 and in Supplementary Table S6A, KEGG terms are often spread across eggNOG. Most of the nodes under human diseases in KEGG appear to be mostly unassigned in eggNOG indicating that as with InterPro:BP, eggNOG is missing medically relevant information available in KEGG.
Table 5.
Matching of KEGG terms to eggNOG terms. For each pair, we report the percentage of reference sequences for a KEGG term that are highly similar to some reference sequence for the corresponding eggNOG term. Matches are shown only for those terms in KEGG that had a best match in eggNOG covering at least 15% of sequences contained in it. The complete list is provided in Supplementary Table S6a
| KEGG | eggNOG | % | |
|---|---|---|---|
| Metabolism | Glycan biosynthesis and metabolism | Cellular processes and signaling → [M] Cell wall/membrane/envelope biogenesis | 55 |
| Not included in regular maps | Metabolism → [C] Energy production and conversion | 63 | |
| Energy metabolism | Metabolism → [C] Energy production and conversion | 40 | |
| Amino acid metabolism | Metabolism → [E] Amino acid transport and metabolism | 56 | |
| Metabolism of other amino acids | Metabolism → [E] Amino acid transport and metabolism | 32 | |
| Biosynthesis of other secondary metabolites | Metabolism → [E] Amino acid transport and metabolism | 34 | |
| Nucleotide metabolism | Metabolism → [F] Nucleotide transport and metabolism | 65 | |
| Carbohydrate metabolism | Metabolism → [G] Carbohydrate transport and metabolism | 35 | |
| Metabolism of cofactors and vitamins | Metabolism → [H] Coenzyme transport and metabolism | 45 | |
| Lipid metabolism | Metabolism → [I] Lipid transport and metabolism | 41 | |
| Metabolism of terpenoids and polyketides | Metabolism → [I] Lipid transport and metabolism | 38 | |
| Xenobiotics biodegradation and metabolism | Metabolism → [I] Lipid transport and metabolism | 19 | |
| Genetic Information Processing | Translation | Information storage and processing → [J] Translation, ribosomal structure and biogenesis | 58 |
| Transcription | Information storage and processing → [K] Transcription | 16 | |
| Replication and repair | Information storage and processing → [L] Replication, recombination and repair | 75 | |
| Folding, sorting and degradation | Cellular processes and signaling → [U] Intracellular trafficking, secretion and vesicular transport | 17 | |
| Environmental information processing | Membrane transport | Metabolism → [P] Inorganic ion transport and metabolism | 25 |
| Signal transduction | Cellular processes and signaling → [T] Signal transduction mechanisms | 18 | |
| Cellular processes | Cell motility | Cellular processes and signaling → [N] Cell motility | 43 |
| Cellular community - prokaryotes | Metabolism → [E] Amino acid transport and metabolism | 27 | |
| Organismal systems | Aging | Cellular processes and signaling → [O] Post-translational modification, protein turnover, chaperones | 38 |
| Human diseases | Drug resistance: antimicrobial | Cellular processes and signaling → [M] Cell wall/membrane/envelope biogenesis | 52 |
| Cardiovascular disease | Cellular processes and signaling → [O] Post-translational modification, protein turnover, chaperones | 28 | |
| Drug resistance: antineoplastic | Cellular processes and signaling → [O] Post-translational modification, protein turnover, chaperones | 16 | |
| Not included in pathway or brite | Poorly characterized | Poorly characterized → [S] Function unknown | 83 |
| Unclassified: genetic information processing | Information storage and processing → [L] Replication, recombination and repair | 47 | |
| Unclassified: metabolism | Poorly characterized → [S] Function unknown | 21 | |
| Unclassified: signaling and cellular processes | Metabolism →[P] Inorganic ion transport and metabolism | 17 | |
The reverse mapping of eggNOG to KEGG (Table 6 and Supplementary Table S6B) shows a similar trend with respect to metabolism and information storage and processing nodes. Except for [B] Chromatin structure and dynamics, hardly any nodes in eggNOG match to Human Diseases in KEGG. Even though KEGG has more sequences than eggNOG, a larger part of eggNOG is unmapped to KEGG than in the opposite direction. This confirms that eggNOG has a greater variety of sequences and is less redundant than KEGG.
Table 6.
Matching of eggNOG terms to KEGG terms. For each pair, we report the percentage of reference sequences for the first term that are highly similar to some reference sequence for the second term. Matches are shown only for those terms in eggNOG that had a best match in KEGG covering at least 15% of sequences contained in it. The complete list is provided in Supplementary Table S6B
| eggNOG | KEGG | % | |
|---|---|---|---|
| Information Storage And Processing | [A] RNA processing and modification | Genetic Information Processing → Translation | 76 |
| [B] Chromatin structure and dynamics | Human Diseases → Cancer: overview | 50 | |
| [J] Translation, ribosomal structure and biogenesis | Genetic Information Processing → Translation | 50 | |
| [L] Replication, recombination and repair | Genetic Information Processing → Replication and repair | 34 | |
| Cellular Processes And Signaling | [M] Cell wall/membrane/envelope biogenesis | Metabolism → Glycan biosynthesis and metabolism | 24 |
| [N] Cell motility | Cellular Processes → Cell motility | 59 | |
| [T] Signal transduction mechanisms | Environmental Information Processing → Signal transduction | 47 | |
| [U] Intracellular trafficking, secretion, and vesicular transport | Environmental Information Processing → Membrane transport | 36 | |
| [V] Defense mechanisms | Environmental Information Processing → Membrane transport | 44 | |
| Metabolism | [C] Energy production and conversion | Metabolism → Energy metabolism | 34 |
| [E] Amino acid transport and metabolism | Metabolism → Amino acid metabolism | 47 | |
| [F] Nucleotide transport and metabolism | Metabolism → Nucleotide metabolism | 75 | |
| [G] Carbohydrate transport and metabolism | Metabolism → Carbohydrate metabolism | 43 | |
| [H] Coenzyme transport and metabolism | Metabolism → Metabolism of cofactors and vitamins | 69 | |
| [I] Lipid transport and metabolism | Metabolism → Lipid metabolism | 44 | |
| [P] Inorganic ion transport and metabolism | Environmental Information Processing → Membrane transport | 28 | |
| Poorly Characterized | [S] Function unknown | Not Included in Pathway or Brite → Poorly characterized | 16 |
Comparison of SEED and eggNOG
Out of all mappings that we considered here, SEED to eggNOG has the best and most meaningful matches. See Table 7 for the best matches from SEED to eggNOG (and Supplementary Table S7A for all matches). SEED appears to have a more user-friendly structure with terms that are slightly more specific than in eggNOG, but that are still general enough to provide a good overview of the data.
Table 7.
Matching of SEED terms to eggNOG terms. For each pair, we report the percentage of reference sequences for the first term that are highly similar to some reference sequence for the second terms. Matches are shown only for those terms in SEED that had a best match in eggNOG covering at least 15% of sequences contained in it. The complete list is provided in Supplementary Table S7A
| SEED | eggNOG | % | |
|---|---|---|---|
| RNA processing | RNA processing and modification | Information storage and processing → [J] Translation, ribosomal structure and biogenesis | 39 |
| Transcription | Information storage and processing → [K] Transcription | 87 | |
| Experimental subsystems | Information storage and processing → [L] Replication, recombination and repair | 75 | |
| DNA processing | DNA recombination | Information storage and processing → [L] Replication, recombination and repair | 58 |
| DNA repair | Information storage and processing → [L] Replication, recombination and repair | 84 | |
| DNA replication | Information storage and processing → [L] Replication, recombination and repair | 90 | |
| DNA sulfur modification | Metabolism → [P] Inorganic ion transport and metabolism | 21 | |
| DNA uptake, competence | Information storage and processing → Replication, recombination and repair | 52 | |
| Type I restriction-modification systems | Cellular processes and signaling → [V] Defense mechanisms | 95 | |
| Protein processing | Protein Synthesis | Information storage and processing → [J] Translation, ribosomal structure and biogenesis | 81 |
| Protein degradation | Cellular processes and signaling → [O] Posttranslational modification, protein turnover, chaperones | 42 | |
| Protein folding | Cellular processes and signaling → [O] Posttranslational modification, protein turnover, chaperones | 94 | |
| Protein processing and modification | Cellular processes and signaling → [O] Posttranslational modification, protein turnover, chaperones | 34 | |
| Protein targeting, sorting, translocation | Cellular processes and signaling → [U] Intracellular trafficking, secretion, and vesicular transport | 55 | |
| Protein glycosylation in prokaryotes | Cellular processes and signaling → [M] Cell wall/membrane/envelope biogenesis | 52 | |
| Selenoproteins | Metabolism → [E] Amino acid transport and metabolism | 32 | |
| Cellular processes | Motility and chemotaxis | Cellular processes and signaling → [N] Cell motility | 74 |
| Microbial communities | Metabolism → [G] Carbohydrate transport and metabolism | 32 | |
| Cell cycle, cell division and death | Cellular processes and signaling → [M] Cell wall/membrane/envelope biogenesis | 27 | |
| Cell envelope | Cellular processes and signaling → [M] Cell wall/membrane/envelope biogenesis | 69 | |
| Energy | Energy and precursor metabolites generation | Metabolism → [C] Energy production and conversion | 44 |
| Respiration | Metabolism → [C] Energy production and conversion | 64 | |
| Clustering-based subsystems | Cell division | Information storage and processing → [K] Transcription | 23 |
| Lysine biosynthesis | Metabolism → [E] Amino acid transport and metabolism | 86 | |
| Possible heterocyst differentiation related cluster | Cellular processes and signaling → [M] Cell wall/membrane/envelope biogenesis | 30 | |
| Shikimate kinase SK3 cluster | Poorly characterized → [S] Function unknown | 40 | |
| Threonine synthase cluster | Metabolism → [E] Amino acid transport and metabolism | 52 | |
| Metabolism | Metabolite damage and its repair or mitigation | Metabolism → [F] Nucleotide transport and metabolism | 27 |
| Nitrogen metabolism | Metabolism → [C] Energy production and conversion | 48 | |
| Amino acids and derivatives | Metabolism → [E] Amino acid transport and metabolism | 73 | |
| Nucleosides and nucleotides | Metabolism → [F] Nucleotide transport and metabolism | 73 | |
| Carbohydrates | Metabolism → [G] Carbohydrate transport and metabolism | 42 | |
| Cofactors, vitamins, prosthetic groups | Metabolism → [H] Coenzyme transport and metabolism | 50 | |
| Fatty acids, lipids and isoprenoids | Metabolism → [I] Lipid transport and metabolism | 53 | |
| Iron acquisition and metabolism | Metabolism → [P] Inorganic ion transport and metabolism | 52 | |
| Phosphate Metabolism | Metabolism → [P] Inorganic ion transport and metabolism | 68 | |
| Sulfur Metabolism | Metabolism → [P] Inorganic ion transport and metabolism | 27 | |
| Regulation and cell signaling | Cellular Processes And Signaling → [T] Signal transduction mechanisms | 49 | |
| Membrane transport | Metabolism → [P] Inorganic ion transport and metabolism | 29 | |
| Secondary metabolism | Metabolism → [Q] Secondary metabolites biosynthesis, transport and catabolism | 23 | |
| Miscellaneous | Poorly Characterized → [S] Function unknown | 36 | |
The reverse mapping as shown in the Table 8 is not as good (see also Supplementary Table S7B). With only five nodes having best matches covering at least 50% of sequences. This is in line with the large redundancy in SEED reported in Figure 2.
Table 8.
Matching of eggNOG terms to SEED terms. For each pair, we report the percentage of reference sequences for the first term that are highly similar to some reference sequence for the second term. Matches are shown only for those terms in eggNOG that had a best match in SEED covering at least 15% of sequences contained in it. The complete list is provided in Supplementary Table S7B
| eggNOG | SEED | % | |
|---|---|---|---|
| Information storage and processing | [A] RNA processing and modification | RNA processing → RNA processing and modification | 75 |
| [B] Chromatin structure and dynamics | Regulation and cell signaling | 69 | |
| [J] Translation, ribosomal structure and biogenesis | Protein processing → Protein synthesis | 62 | |
| [L] Replication, recombination and repair | DNA processing → DNA repair | 29 | |
| Cellular processes and signaling | [D] Cell cycle control, cell division, chromosome partitioning | Cellular processes → Cell cycle, cell division and death | 34 |
| [M] Cell wall/membrane/envelope biogenesis | Cell envelope | 15 | |
| [N] Cell motility | Motility and chemotaxis | 60 | |
| [O] Post-translational modification, protein turnover, chaperones | Protein processing → Protein folding | 16 | |
| [U] Intracellular trafficking, secretion and vesicular transport | Membrane transport | 29 | |
| [V] Defense mechanisms | Membrane transport | 15 | |
| Metabolism | [C] Energy production and conversion | Energy → Respiration | 23 |
| [E] Amino acid transport and metabolism | Metabolism → Amino acids and derivatives | 40 | |
| [F] Nucleotide transport and metabolism | Metabolism → Nucleosides and nucleotides | 43 | |
| [G] Carbohydrate transport and metabolism | Metabolism → Carbohydrates | 26 | |
| [H] Coenzyme transport and metabolism | Metabolism → Cofactors, vitamins, prosthetic groups | 65 | |
| [I] Lipid transport and metabolism | Metabolism → Fatty acids, lipids, and isoprenoids | 35 | |
| [P] Inorganic ion transport and metabolism | Membrane Transport | 24 | |
| [Q] Secondary metabolites biosynthesis, transport and catabolism | Metabolism → Cofactors, vitamins, prosthetic groups | 18 | |
Comparison of SEED and KEGG
As expected, we obtained a much better mapping of SEED to KEGG than vice versa. As shown in Figure 3 and Supplementary Table S8B, a large proportion of the nodes in KEGG can not be mapped at all. Nodes with matches are listed in Table 10. There is a correspondence between metabolism nodes in both classification systems, however other terms appear to be quite scattered. Matches of SEED nodes to Organismal Systems and Human Diseases in KEGG are very weak, indicating that KEGG is better suited than SEED for analyses in a medical context, see Table 9.
Table 10.
Matching of KEGG terms to SEED terms. For each pair, we report the percentage of reference sequences for a KEGG node that are highly similar to some reference sequence for the corresponding eggNOG node. Matches are shown only for those nodes in KEGG that had a best match in SEED covering at least 15% of sequences contained in it. The complete list is provided in Supplementary Table S6A
| KEGG | SEED | % | |
|---|---|---|---|
| Metabolism | Glycan biosynthesis and metabolism | Cell envelope | 26 |
| Not included in regular maps | Metabolism → Amino acids and derivatives | 28 | |
| Energy metabolism | Energy → Respiration | 24 | |
| Amino acid metabolism | Metabolism → Amino acids and derivatives | 47 | |
| Metabolism of other amino acids | Metabolism → Amino acids and derivatives | 17 | |
| Biosynthesis of other secondary metabolites | Metabolism → Amino acids and derivatives | 25 | |
| Nucleotide metabolism | Metabolism → Nucleosides and nucleotides | 40 | |
| Carbohydrate metabolism | Energy → Energy and precursor metabolites generation | 32 | |
| Metabolism of cofactors and vitamins | Metabolism → Cofactors, vitamins, prosthetic groups | 52 | |
| Lipid metabolism | Metabolism → Fatty acids, lipids and isoprenoids | 35 | |
| Metabolism of terpenoids and polyketides | Metabolism → Fatty acids, lipids and isoprenoids | 28 | |
| Genetic Information Processing | Translation | Protein processing → Protein synthesis | 62 |
| Transcription | RNA processing → Transcription | 22 | |
| Replication and repair | DNA processing → DNA repair | 47 | |
| Environmental Information Processing | Membrane transport | Membrane transport | 21 |
| Cellular Processes | Cell motility | Cellular processes → Motility and chemotaxis | 52 |
| Cellular community - prokaryotes | Membrane transport | 28 | |
| Organismal Systems | Aging | Stress response, defense and virulence | 25 |
| Human Diseases | Drug resistance: antimicrobial | Membrane transport | 32 |
| Cardiovascular disease | Stress response, defense and virulence | 15 | |
| Drug resistance: antineoplastic | Stress response, defense and virulence | 16 | |
Table 9.
Matching of SEED terms to KEGG terms. For each pair, we report the percentage of reference sequences for the first term that are highly similar to some reference sequence for the second term. Matches are shown only for those terms in SEED that had a best match in KEGG covering at least 15% of sequences contained in it. The complete list is provided in Supplementary Table S8A
| SEED | KEGG | % | |
|---|---|---|---|
| RNA Processing | RNA processing and modification | Genetic information processing → Folding, sorting and degradation | 24 |
| Transcription | Genetic information processing → Transcription | 93 | |
| DNA Processing | DNA recombination | Genetic information processing → Replication and repair | 40 |
| DNA repair | Genetic information processing → Replication and repair | 68 | |
| DNA replication | Human diseases → Drug resistance: antineoplastic | 22 | |
| DNA uptake, competence | Genetic information processing → Replication and repair | 44 | |
| Protein Processing | Protein synthesis | Genetic information processing → Translation | 64 |
| Protein degradation | Cellular processes → Cell growth and death | 23 | |
| Protein folding | Genetic information processing → Folding, sorting and degradation | 31 | |
| Protein processing and modification | Not included in pathway or brite → Unclassified: metabolism | 35 | |
| Protein targeting, sorting, translocation | Genetic information processing → Folding, sorting and degradation | 56 | |
| Protein glycosylation in Prokaryotes | Not included in pathway or Brite → Unclassified: metabolism | 41 | |
| Selenoproteins | Metabolism → Metabolism of other amino acids | 34 | |
| Cellular Processes | Motility and Chemotaxis | Cellular Processes → Cell motility | 92 |
| Microbial communities | Cellular Processes → Cellular community - prokaryotes | 84 | |
| Cell cycle, cell division and death | Metabolism → Glycan biosynthesis and metabolism | 21 | |
| Prokaryotic cell type differentiation | Not Included in Pathway or Brite → Unclassified: signaling and cellular processes | 58 | |
| Cell Envelope | Metabolism → Glycan biosynthesis and metabolism | 59 | |
| Energy | Energy and precursor metabolites generation | Metabolism → Carbohydrate metabolism | 78 |
| Photosynthesis | Metabolism → Energy metabolism | 57 | |
| Respiration | Metabolism → Energy metabolism | 58 | |
| Clustering-based subsystems | Cell division | Metabolism → Metabolism of cofactors and vitamins | 15 |
| Lysine biosynthesis | Metabolism → Amino acid metabolism | 85 | |
| Possible heterocyst differentiation related cluster | Not Included in Pathway or Brite → Unclassified: signaling and cellular processes | 30 | |
| Shikimate kinase SK3 cluster | Not Included in Pathway or Brite → Poorly characterized | 35 | |
| Threonine synthase cluster | Metabolism → Amino acid metabolism | 56 | |
| Metabolism | Metabolite damage and its repair or mitigation | Metabolism → Nucleotide metabolism | 35 |
| Nitrogen metabolism | Metabolism → Energy metabolism | 55 | |
| Amino acids and derivatives | Metabolism → Amino acid metabolism | 76 | |
| Nucleosides and nucleotides | Metabolism → Nucleotide metabolism | 78 | |
| Carbohydrates | Metabolism → Carbohydrate metabolism | 55 | |
| Cofactors, vitamins, prosthetic groups | Metabolism → Metabolism of cofactors and vitamins | 62 | |
| Fatty acids, lipids and isoprenoids | Metabolism → Lipid metabolism | 57 | |
| Iron acquisition and metabolism | Environmental information processing → Membrane transport | 33 | |
| Phosphate metabolism | Environmental Information Processing → Membrane transport | 52 | |
| Sulfur metabolism | Metabolism → Energy metabolism | 54 | |
| Regulation and cell signaling | Metabolism → Nucleotide metabolism | 18 | |
| Membrane transport | Environmental information processing → Membrane transport | 25 | |
| Secondary metabolism | Metabolism → Amino acid metabolism | 46 | |
| Miscellaneous | Not included in pathway or brite → Unclassified: metabolism | 18 | |
Self mappings
How well does each functional classification system map to itself? To investigate this, we determined how many sequences are shared among nodes of the same functional classification system. Apart from the expected perfect matches along the diagonals, as shown in Figure 3 and Supplementary Tables S13A–D, we also find some overlaps among different nodes. In InterPro:BP this is explained by the fact that the classification is constructed from a directed acyclic graph and is not based on a hierarchy. This leads to some nodes appearing multiple times on the tree and thus duplicated occurrences of the same terms and sequences.
The eggNOG classification has very few pairs of terms or nodes that ‘intersect’, that is, share reference sequences. For example, [W] Extracellular structures are completely contained in [U] Intracellular trafficking, secretion and vesicular transport. In fact [W] Extracellular structures contain only one COG (COG5295 domain protein) that is also contained in the [U].
While KEGG is hierarchically structured, the same KEGG orthology groups may appear in different metabolic pathways, giving rise to sequences that are associated with multiple nodes. For the two groups ‘Metabolism’ and ‘Environmental Information Processing, Cellular Processes, Organismal Systems and Human Diseases’, that are a lot of intersections between nodes in the same group.
For SEED, all intersections among nodes are empty, indicating a strictly hierarchical classification.
Illustration of classifications on short read data
When it comes to functional classification of reads there are two questions of interest. The first question is: how do functional classification systems differ with respect to the number of reads that can be mapped to them? In other words, to what extend do functional classification systems overlap and complement each other when it comes to actual data analysis. To explore this, we considered all four large databases and counted reads with significant matches.
The results are reported in Figure 4. There are two patterns. In the first two listed pairs of functional classifications, both classifications roughly covered each other. InterPro:BP shares most of the assigned reads with eggNOG and KEGG with SEED. In the other cases, the first classification complements the other but not vice versa. This is true for all pairs that involve KEGG or SEED but not both. Here about half of the reads that can be mapped to either of the two, have matches in KEGG or SEED whereas almost all mapped reads have matches in either InterPro:BP or eggNOG.
Figure 4.
For each pair of functional classifications A and B listed on the left, and for five different metagenomic sequencing samples (labeled Gut, Skin, Water, Landfill 1 and Landfill 2), we display a bar chart indicating the proportion of reads assigned only in classification A (left), in both A and B (middle), and only in B (right), respectively.
The second question is how sensitive is a functional analysis, as performed by MEGAN, say, on the output of a DIAMOND alignment of a metagenomic sample against the whole NCBI-nr database, in comparison to such an analysis performed on alignments against a special-purpose database that only contains reference sequences for a specific functional classification.
To address this, we analyzed all five metagenome samples, using all four main functional classifications, twice each, once based on an alignment of all reads against the NCBI-nr database, and once based on an alignment of all reads against a database containing all references sequences for the functional classification. Functional analysis was performed using MEGAN (default parameters). The 5 × 4 × 2 results are plotted in Figure 5. In all cases, alignment against a specific database gave rise to substantially more mapped reads than alignment against NCBI-nr. Also, more reads could be assigned to InterPro:BP than to the other other three classifications, and more to eggNOG than to KEGG or SEED.
Figure 5.
Percentage of reads mapped to functional classification databases with the basic DIAMOND+MEGAN pipeline. Alignment was performed either against a classification specific database (labeled specific) or the whole of NCBI-nr (labeled nr).
Based on these comparisons, the databases can be ranked as follows: InterPro:BP > eggNOG > KEGG > SEED.
Taxonomic composition of functional databases
The above analysis suggests that mappings of SEED and KEGG to eggNOG have much lower assignment rates that to InterPro:BP. However, in practical applications on metagenomic data eggNOG appears to perform almost as well as InterPro:BP. To explore this, we looked into taxonomic composition of all four functional classification systems. Figure 6 shows how sequences within each database are distributed among archeae, bacteria, eukarya and viruses. A large proportion of sequences in all databases are bacterial, however eggNOG is almost exclusively bacterial (99.7% of all sequences). SEED contains sequences that belong to unclassified organisms that are probably bacteria, whereas KEGG and InterPro:BP include sequences from eukaryotic organisms and viruses.
Figure 6.
Sequence distribution with respect to taxonomy in the four large functional classification systems.
Next we looked at which proportion of bacterial sequences in InterPro:BP, KEGG and SEED could have been mapped to eggNOG and compared it to the proportion of total sequences that have matches in it. As shown in the Table 3, 68.1% of all sequences in InterPro:BP, 75.4% in KEGG and 97.3% in SEED have matches in eggNOG. Table 11 shows which proportion of sequences that belong to different kingdoms have matches in eggNOG. Whereas the number of matched bacterial sequences in SEED is almost the same as for all sequences, it is much higher in KEGG and InterPro:BP. Hence, eggNOG is just as well suited for microbiome analyses as larger functional classification systems are.
Table 11.
Percentage of sequences originating from different Kingdoms in InterPro:BP, KEGG and SEED that have matches in eggNOG
| InterPro:BP | KEGG | SEED | |
|---|---|---|---|
| Unclassified | 63.4 | 73.2 | 97.4 |
| Archaea | 67.5 | 76.2 | 77.2 |
| Bacteria | 91.4 | 98.2 | 97.5 |
| Eukarya | 41.9 | 63.8 | 93.6 |
| Virus | 4.5 | 80.0 | 82.8 |
To explore whether medically relevant nodes in KEGG and InterPro:BP are of interest for metagenomic studies, we looked at the taxonomic composition of all nodes in the two as considered in our comparisons. Figure 7 shows taxonomic composition for KEGG. Except ‘Drug resistance: antimicrobial’, all other medically relevant terms under organismal systems and human diseases have more sequences from Eukaryotic organisms than from Bacteria, however the amount of bacterial sequences makes those two categories just as relevant for metagenomics research as other KEGG terms. Regarding medically relevant nodes in InterPro:BP, none of them is dominated by eukaryotic sequences as shown in the Figure 8.
Figure 7.
Proportions of bacterial, eukaryotic and other sources of sequences for nodes in KEGG that were used in comparisons against other functional classification systems.
Figure 8.
Proportions of bacterial, eukaryotic and other sources of sequences for nodes in InterPro:BP that were used in comparisons against other functional classification systems.
Small specific databases
All three antibiotic resistance databases map well to eggNOG → [V] Defense mechanisms, KEGG → Drug resistance: antimicrobial, KEGG → Protein families: signaling and cellular processes and SEED → Stress Response, Defense, Virulence. In InterPro:BP the mappings are not so specific. Most of the sequences can be assigned to GO:0008152 metabolic process, GO:0006810 transport and GO:0042221 response to chemical.
The metabolic databases Enzyme and MetaCyc are spread across nodes under eggNOG → Metabolism, KEGG → Metabolism, SEED → Metabolism and GO:0008152 metabolic process.
VFDB maps best to nodes associated with cell membranes, cell motility and intracellular transport in all large functional classification systems.
Out of the seven small databases considered here, VFDB core and VFDB full have the largest proportion of unmapped sequences in eggNOG, KEGG and SEED. InterPro:BP contains matching sequences for almost all of the small databases with only 1–4% of unmatched sequences per database. SEED on the other hand performs worst with 19% to 54% of missing matches. See Supplementary Tables S9–S12 for more details.
DISCUSSION
In this paper we considered a variety of aspects for analysing and comparing functional classification systems relevant for metagenomic analyses. The size of the database proved not to be a reliable indicator for its functionality. Clustering at 90% and 70% sequence identity shows that the three largest databases (InterPro:BP, SEED and KEGG) are much more redundant than eggNOG, which is smallest of the four. This explains why we were able to map more reads to eggNOG than to SEED or KEGG, even though both are larger than eggNOG.
We noticed that the four large functional classification systems differ much more in their hierarchical structure than in their sequence content. The term Metabolism appears in eggNOG, KEGG and SEED and can be mapped well among the three. However, the term Metabolic Process in InterPro:BP is much more general and does not map as well. In general we found that terms in InterPro:BP are either too general or too specific and thus not well suited for providing an overview. In contrast, eggNOG provides a hierarchy with evenly distributed sequences and provides the best structure for the amount of sequence content. KEGG and SEED are also well structured, however they appear to contain less information with fewer read assignments than eggNOG or InterPro:BP.
Even though we would recommend using eggNOG for general data overviews, it might be less useful when it comes to medical applications. Medically relevant KEGG terms are spread across nodes in eggNOG whereas such terms in InterPro:BP remain largely unassigned. Therefore KEGG and InterPro:BP might be more informative for medical applications.
There are other aspects that we have not considered here that might play a role when choosing a functional classification system. For example, the visualizations of metabolic pathways provided by KEGG may help to analyze the metabolic potential of a sample.
Small specific databases are mostly contained within the four larger ones and can be mapped to the nodes with names that reflect their function. This, however, does not mean that general databases could replace specialized ones. For one, they posses different hierarchical structures that may be more suitable for answering various questions. Furthermore, as shown in the examples in the Figure 5, due to the limited number of hits reported by a homology search, it is more likely that more reads get meaningful assignments when a small specific database is chosen rather than a large one. On the other hand, using a small database increases chances of false positive hits, which might lead to inaccurate interpretations.
DATA AVAILABILITY
All scripts used in this study are available on Code Ocean (DOI 10.24433/CO.3467358.v1) (49). Short read data can be downloaded from Sequence Read Archive: https://www.ncbi.nlm.nih.gov/sra. MEGAN and the MEGAN mapping database (megan-map-Jan2021.db) are available from https://software-ab.informatik.uni-tuebingen.de/download/megan6. All sequence files for small databases are available from their respective websites.
Supplementary Material
ACKNOWLEDGEMENTS
We would like to thank Anupam Gautam for helpful discussions and support. The authors acknowledge infrastructural support by the cluster of Excellence EXC2124 Controlling Microbes to Fight Infection (CMFI), project ID 390838134.
Contributor Information
Monika Zeller, Algorithms in Bioinformatics, Institute for Bioinformatics and Medical Informatics, University of Tübingen, Sand 14, 72076 Tübingen, Germany.
Daniel H Huson, Algorithms in Bioinformatics, Institute for Bioinformatics and Medical Informatics, University of Tübingen, Sand 14, 72076 Tübingen, Germany.
SUPPLEMENTARY DATA
Supplementary Data are available at NARGAB Online.
FUNDING
Zentrum für Datenverarbeitung, University of Tübingen; Baden-Württemberg through bwHPC; German Research Foundation (DFG) [grant number INST 37/935-1 FUGG].
Conflict of interest statement. None declared.
REFERENCES
- 1. Huson D.H., Mitra S., Weber N., Ruscheweyh H.-J., Schuster S.C.. Integrative analysis of environmental sequences using MEGAN 4. Genome Res. 2011; 21:1552–1560. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Glass E.M., Wilkening J., Wilke A., Antonopoulos D., Meyer F.. Using the Metagenomics RAST Server (MG-RAST) for Analyzing Shotgun Metagenomes. Cold Spring Harb Protoc. 2010; 2010:pdb.prot5368. [DOI] [PubMed] [Google Scholar]
- 3. Huerta-Cepas J., Szklarczyk D., Heller D., Hernández-Plaza A., Forslund S.K., Cook H., Mende D.R., Letunic I., Rattei T., Jensen L.J.et al.. eggNOG 5.0: a hierarchical, functionally and phylogenetically annotated orthology resource based on 5090 organisms and 2502 viruses. Nucleic Acids Res. 2018; 47:D309–D314. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Kanehisa M., Furumichi M., Tanabe M., Sato Y., Morishima K.. KEGG: new perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res. 2016; 45:D353–D361. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Mitchell A.L., Attwood T.K., Babbitt P.C., Blum M., Bork P., Bridge A., Brown S.D., Chang H.-Y., El-Gebali S., Fraser M.I.et al.. InterPro in 2019: improving coverage, classification and access to protein sequence annotations. Nucleic Acids Res. 2018; 47:D351–D360. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Overbeek R., Olson R., Pusch G.D., Olsen G.J., Davis J.J., Disz T., Edwards R.A., Gerdes S., Parrello B., Shukla M.et al.. The SEED and the Rapid Annotation of microbial genomes using Subsystems Technology (RAST). Nucleic Acids Res. 2013; 42:D206–D214. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Bairoch A. The ENZYME database in 2000. Nucleic Acids Res. 2000; 28:304–305. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Caspi R., Billington R., Fulcher C.A., Keseler I.M., Kothari A., Krummenacker M., Latendresse M., Midford P.E., Ong Q., Ong W.K.et al.. The MetaCyc database of metabolic pathways and enzymes. Nucleic Acids Res. 2017; 46:D633–D639. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Alcock B.P., Raphenya A.R., Lau T.T.Y., Tsang K.K., Bouchard M., Edalatmand A., Huynh W., Nguyen A.-L.V., Cheng A.A.et al.. CARD 2020: antibiotic resistome surveillance with the comprehensive antibiotic resistance database. Nucleic Acids Res. 2019; 48:D517–D525. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Lakin S.M., Dean C., Noyes N.R., Dettenwanger A., Ross A.S., Doster E., Rovira P., Abdo Z., Jones K.L., Ruiz J.et al.. MEGARes: an antimicrobial resistance database for high throughput sequencing. Nucleic Acids Res. 2016; 45:D574–D580. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Arango-Argoty G.A., Guron G. K.P., Garner E., Riquelme M.V., Heath L.S., Pruden A., Vikesland P.J., Zhang L.. ARGminer: a web platform for the crowdsourcing-based curation of antibiotic resistance genes. Bioinformatics. 2020; 36:2966–2973. [DOI] [PubMed] [Google Scholar]
- 12. Chen L., Yang J., Yu J., Yao Z., Sun L., Shen Y., Jin Q.. VFDB: a reference database for bacterial virulence factors. Nucleic Acids Res. 2005; 33:D325–D328. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Ikram N., Qadir M.A., Afzal M.T.. Investigating correlation between protein sequence similarity and semantic similarity using Gene Ontology Annotations. IEEE/ACM Trans. Comput. Biol. Bioinform. 2018; 15:905–912. [DOI] [PubMed] [Google Scholar]
- 14. Altschul S.F., Gish W., Miller W., Myers E.W., Lipman D.J.. Basic local alignment search tool. J. Mol. Biol. 1990; 215:403–410. [DOI] [PubMed] [Google Scholar]
- 15. Buchfink B., Xie C., Huson D.H.. Fast and sensitive protein alignment using DIAMOND. Nat. Methods. 2015; 12:59–60. [DOI] [PubMed] [Google Scholar]
- 16. Mitchell A.L., Scheremetjew M., Denise H., Potter S., Tarkowska A., Qureshi M., Salazar G.A., Pesseat S., Boland M.A., Hunter F.M.I.et al.. EBI Metagenomics in 2017: enriching the analysis of microbial communities, from sequence reads to assemblies. Nucleic Acids Res. 2018; 46:D726–D735. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Huson D.H., Beier S., Flade I., Górska A., El-Hadidi M., Mitra S., Ruscheweyh H.-J., Tappu R.. MEGAN community edition-interactive exploration and analysis of large-scale microbiome sequencing data. PLoS Comput. Biol. 2016; 12:e1004957. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Ashburner M., Ball C.A., Blake J.A., Botstein D., Butler H., Cherry J.M., Davis A.P., Dolinski K., Dwight S.S., Eppig J.T.et al.. Gene Ontology: tool for the unification of biology. Nat. Genet. 2000; 25:25–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Tatusov R.L., Koonin E.V., Lipman D.J.. A genomic perspective on protein families. Science. 1997; 278:631–637. [DOI] [PubMed] [Google Scholar]
- 20. Overbeek R., Begley T., Butler R.M., Choudhuri J.V., Chuang H.Y., Cohoon M., de Crécy-Lagard V., Diaz N., Disz T., Edwards R.et al.. The subsystems approach to genome annotation and its use in the project to annotate 1000 genomes. Nucleic Acids Res. 2005; 33:5691–702. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Galperin M.Y., Makarova K.S., Wolf Y.I., Koonin E.V.. Expanded microbial genome coverage and improved protein family annotation in the COG database. Nucleic Acids Res. 2015; 43:D261–D269. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Tatusov R., Fedorova N., Jackson J., Jacobs A., Kiryutin B., Koonin E., Krylov D., Mazumder R., Mekhedov S., Nikolskaya A.et al.. The COG database: an updated version includes eukaryotes. BMC Bioinform. 2003; 4:41. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Makarova K.S., Wolf Y.I., Koonin E.V.. Archaeal clusters of orthologous genes (arCOGs): an update and application for analysis of shared features between thermococcales, methanococcales, and methanobacteriales. Life. 2015; 5:818–840. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Kanehisa M., Sato Y., Furumichi M., Morishima K., Tanabe M.. New approach for understanding genome variations in KEGG. Nucleic Acids Res. 2018; 47:D590–D595. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Lewis T.E., Sillitoe I., Dawson N., Lam S.D., Clarke T., Lee D., Orengo C., Lees J.. Gene3D: Extensive prediction of globular domains in proteins. Nucleic Acids Res. 2017; 46:D435–D439. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Marchler-Bauer A., Bo Y., Han L., He J., Lanczycki C.J., Lu S., Chitsaz F., Derbyshire M.K., Geer R.C., Gonzales N.R.et al.. CDD/SPARCLE: functional classification of proteins via subfamily domain architectures. Nucleic Acids Res. 2016; 45:D200–D203. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Pedruzzi I., Rivoire C., Auchincloss A.H., Coudert E., Keller G., de Castro E., Baratin D., Cuche B.A., Bougueleret L., Poux S.et al.. HAMAP in 2015: updates to the protein family classification and annotation system. Nucleic Acids Res. 2014; 43:D1064–D1070. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Mi H., Huang X., Muruganujan A., Tang H., Mills C., Kang D., Thomas P.D.. PANTHER version 11: expanded annotation data from Gene Ontology and Reactome pathways, and data analysis tool enhancements. Nucleic Acids Res. 2016; 45:D183–D189. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Finn R.D., Coggill P., Eberhardt R.Y., Eddy S.R., Mistry J., Mitchell A.L., Potter S.C., Punta M., Qureshi M., Sangrador-Vegas A.et al.. The Pfam protein families database: towards a more sustainable future. Nucleic Acids Res. 2015; 44:D279–D285. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Nikolskaya A.N., Arighi C.N., Huang H., Barker W.C., Wu C.H.. PIRSF family classification system for protein functional and evolutionary analysis. Evol. Bioinform. 2006; 2:117693430600200033. [PMC free article] [PubMed] [Google Scholar]
- 31. Attwood T.K., Coletta A., Muirhead G., Pavlopoulou A., Philippou P.B., Popov I., Romá-Mateo C., Theodosiou A., Mitchell A.L.. The PRINTS database: a fine-grained protein sequence annotation and analysis resource – its status in 2012. Database. 2012; 2012:bas019. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Bru C., Courcelle E., Carrè S., Beausse Y., Dalmar S., Kahn D.. The ProDom database of protein domain families: more emphasis on 3D. Nucleic Acids Res. 2005; 33:D212–D215. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Sigrist C.J.A., de Castro E., Cerutti L., Cuche B.A., Hulo N., Bridge A., Bougueleret L., Xenarios I.. New and continuing developments at PROSITE. Nucleic Acids Res. 2012; 41:D344–D347. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. Letunic I., Bork P.. 20 years of the SMART protein domain annotation resource. Nucleic Acids Res. 2017; 46:D493–D496. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Akiva E., Brown S., Almonacid D.E., Barber A.E., Custer A.F., Hicks M.A., Huang C.C., Lauck F., Mashiyama S.T., Meng E.C., Mischel D.et al.. The Structure–Function Linkage Database. Nucleic Acids Res. 2013; 42:D521–D530. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Oates M.E., Stahlhacke J., Vavoulis D.V., Smithers B., Rackham O.J., Sardar A.J., Zaucha J., Thurlby N., Fang H., Gough J.. The SUPERFAMILY 1.75 database in 2014: a doubling of data. Nucleic Acids Res. 2014; 43:D227–D233. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37. Haft D.H., Selengut J.D., Richter R.A., Harkins D., Basu M.K., Beck E.. TIGRFAMs and Genome Properties in 2013. Nucleic Acids Res. 2012; 41:D387–D395. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Aziz R.K., Bartels D., Best A.A., DeJongh M., Disz T., Edwards R.A., Formsma K., Gerdes S., Glass E.M., Kubal M.et al.. The RAST Server: Rapid Annotations using Subsystems Technology. BMC Genom. 2008; 9:75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Henry C.S., DeJongh M., Best A.A., Frybarger P.M., Linsay B., Stevens R.L.. High-throughput generation, optimization and analysis of genome-scale metabolic models. Nat. Biotechnol. 2010; 28:977–982. [DOI] [PubMed] [Google Scholar]
- 40. Devoid S., Overbeek R., DeJongh M., Vonstein V., Best A.A., Henry C.. Alper H.S. Automated genome annotation and metabolic model reconstruction in the SEED and Model SEED. Systems Metabolic Engineering:Methodsand Protocols.Methods in Molecular Biology. 2013; 985:Totowa, NJ: Humana Press; 17–45. [DOI] [PubMed] [Google Scholar]
- 41. Canese K., Weis S.. PubMed: the bibliographic database. The NCBI Handbook [Internet]. 2013; 2nd editionBethesda, MD: National Center for Biotechnology Information (US)https://www.ncbi.nlm.nih.gov/books/NBK153385/. [Google Scholar]
- 42. Wattam A.R., Davis J.J., Assaf R., Boisvert S., Brettin T., Bun C., Conrad N., Dietrich E.M., Disz T., Gabbard J.L.et al.. Improvements to PATRIC, the all-bacterial Bioinformatics Database and Analysis Resource Center. Nucleic Acids Res. 2016; 45:D535–D542. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43. Webb E.C.et al.. Enzyme nomenclature 1992. Recommendations of the Nomenclature Committee of the International Union of Biochemistry and Molecular Biology on the Nomenclature and Classification of Enzymes. 1992; San Diego: Academic Press; Number Ed. 6. [Google Scholar]
- 44. Yang Y., Jiang X., Chai B., Ma L., Li B., Zhang A., Cole J.R., Tiedje J.M., Zhang T.. ARGs-OAP: online analysis pipeline for antibiotic resistance genes detection from metagenomic data using an integrated structured ARG-database. Bioinformatics. 2016; 32:2346–2351. [DOI] [PubMed] [Google Scholar]
- 45. Li W., Godzik A.. Cd-hit: a fast program for clustering and comparing large sets of protein or nucleotide sequences. Bioinformatics. 2006; 22:1658–1659. [DOI] [PubMed] [Google Scholar]
- 46. Edgar R.C. Search and clustering orders of magnitude faster than BLAST. Bioinformatics. 2010; 26:2460–2461. [DOI] [PubMed] [Google Scholar]
- 47. Leinonen R., Sugawara H., Shumway M.. The Sequence Read Archive. Nucleic Acids Res. 2010; 39:D19–D21. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48. Chen S., Zhou Y., Chen Y., Gu J.. fastp: an ultra-fast all-in-one FASTQ preprocessor. Bioinformatics. 2018; 34:i884–i890. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Zeller M., Huson D.H.. Comparison of Functional Classification Systems [Source Code]. 2022; (22 August 2022, date last accessed) 10.24433/CO.3467358.v1. [DOI] [PMC free article] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All scripts used in this study are available on Code Ocean (DOI 10.24433/CO.3467358.v1) (49). Short read data can be downloaded from Sequence Read Archive: https://www.ncbi.nlm.nih.gov/sra. MEGAN and the MEGAN mapping database (megan-map-Jan2021.db) are available from https://software-ab.informatik.uni-tuebingen.de/download/megan6. All sequence files for small databases are available from their respective websites.