A unique and customizable approach for functionally annotating prokaryotic genomes
MacSyFinder v2: Improved modelling and search engine to identify molecular systems in genomes
Recommendation: posted 17 February 2023, validated 24 February 2023
Macromolecular System Finder (MacSyFinder) v2 (Néron et al., 2023) is a newly updated approach for performing functional annotation of prokaryotic genomes (Abby et al., 2014). This tool parses an input file of protein sequences from a single genome (either ordered by genome location or unordered) and identifies the presence of specific cellular functions (referred to as “systems”). These systems are called based on two criteria: (1) that the "quorum" of a minimum set of core proteins involved is reached the “quorum” of a minimum set of core proteins being involved that are present, and (2) that the genes encoding these proteins are in the expected genomic organization (e.g., within the same order in an operon), when ordered data is provided. I believe the MacSyFinder approach represents an improvement over more commonly used methods exactly because it can incorporate such information on genomic organization, and also because it is more customizable.
Before properly appreciating these points, it is worth noting the norms and key challenges surrounding high-throughput functional annotation of prokaryotic genomes. Genome sequences are being added to online repositories at increasing rates, which has led to an enormous amount of bacterial genome diversity available to investigate (Altermann et al., 2022). A key aspect of understanding this diversity is the functional annotation step, which enables genes to be grouped into more biologically interpretable categories. For instance, gene calls can be mapped against existing Clusters of Orthologous Genes, which are themselves grouped into general categories such as ‘Transcription’ and ‘Lipid metabolism’ (Galperin et al., 2021).
This approach is valuable but is primarily used for global summaries of functional annotations within a genome: for example, it could be useful to know that a genome is particularly enriched for genes involved in lipid metabolism. However, knowing that a particular gene is involved in the general process of lipid metabolism is less likely to be actionable. In other words, the desired specificity of a gene’s functional annotation will depend on the exact question being investigated. There is no shortage of functional ontologies in genomics that can be applied for this purpose (Douglas and Langille, 2021), and researchers are often overwhelmed by the choice of which functional ontology to use. In this context, giving researchers the ability to precisely specify the gene families and operon structures they are interested in identifying across genomes provides useful control over what precise functions they are profiling. Of course, most researchers will lack the information and/or expertise to fully take advantage of MacSyFinder’s customizable features, but having this option for specialized purposes is valuable.
The other MacSyFinder feature that I find especially noteworthy is that it can incorporate genomic organization (e.g., of genes ordered in operons) when calling systems. This is a rare feature among commonly used tools for functional annotation and likely results in much higher specificity. As the authors note, this capability makes the co-occurrence of paralogs, and other divergent genes that share sequence similarity, to contribute less noise (i.e., they result in fewer false positive calls).
It is important to emphasize that these features are not new additions in MacSyFinder v2, but there are many other valuable changes. Most practically, this release is written in Python 3, rather than the obsolete Python 2.7, and was made more computationally efficient, which will enable MacSyFinder to be more widely used and more easily maintained moving forward. In addition, the search algorithm for analyzing individual proteins was fundamentally updated as well. The authors show that their improvements to the search algorithm result in an 8% and 20% increase in the number of identified calls for single and multi-locus secretion systems, respectively. Taken together, MacSyFinder v2 represents both practical and scientific improvements over the previous version, which will be of great value to the field.
Abby SS, Néron B, Ménager H, Touchon M, Rocha EPC (2014) MacSyFinder: A Program to Mine Genomes for Molecular Systems with an Application to CRISPR-Cas Systems. PLOS ONE, 9, e110726. https://doi.org/10.1371/journal.pone.0110726
Altermann E, Tegetmeyer HE, Chanyi RM (2022) The evolution of bacterial genome assemblies - where do we need to go next? Microbiome Research Reports, 1, 15. https://doi.org/10.20517/mrr.2022.02
Douglas GM, Langille MGI (2021) A primer and discussion on DNA-based microbiome data and related bioinformatics analyses. Peer Community Journal, 1. https://doi.org/10.24072/pcjournal.2
Galperin MY, Wolf YI, Makarova KS, Vera Alvarez R, Landsman D, Koonin EV (2021) COG database update: focus on microbial diversity, model organisms, and widespread pathogens. Nucleic Acids Research, 49, D274–D281. https://doi.org/10.1093/nar/gkaa1018
Néron B, Denise R, Coluzzi C, Touchon M, Rocha EPC, Abby SS (2023) MacSyFinder v2: Improved modelling and search engine to identify molecular systems in genomes. bioRxiv, 2022.09.02.506364, ver. 2 peer-reviewed and recommended by Peer Community in Genomics. https://doi.org/10.1101/2022.09.02.506364
Gavin Douglas (2023) A unique and customizable approach for functionally annotating prokaryotic genomes. Peer Community in Genomics, 100233. https://doi.org/10.24072/pci.genomics.100233
The recommender in charge of the evaluation of the article and the reviewers declared that they have no conflict of interest (as defined in the code of conduct of PCI) with the authors or with the content of the article. The authors declared that they comply with the PCI rule of having no financial conflicts of interest in relation to the content of the article.
EPCR lab acknowledges funding from the INCEPTION project (ANR-16-CONV-0005), Equipe FRM (Fondation pour la Recherche Médicale): EQU201903007835, and Laboratoire d’Excellence IBEID Integrative Biology of Emerging Infectious Diseases (ANR-10-LABX-62-IBEID). SSA received financial support from the CNRS and TIMC lab (INSIS “starting grant”) and the French National Research Agency, “Investissements d’avenir” program ANR-15-IDEX-02.
Evaluation round #1
DOI or URL of the preprint: https://www.biorxiv.org/content/10.1101/2022.09.02.506364v1
Author's Reply, 22 Jan 2023
Decision by Gavin Douglas, posted 12 Oct 2022
To Dr. Abby and colleagues,
Two reviewers have examined your manuscript and they both agree it is interesting and is of value for the field, which I also agree with. Their overlapping comments pertain to clarifying certain points of confusion, through adding additional text and figures where needed, and also clarifying how metagenome-assembled genomes could be used as input.
I also have some comments in addition to what the reviewers brought up, mainly related to adding additional clarification. Please see below.
I think after addressing all of our combined comments that your manuscript will be clearer, particularly for users new to the MacSyFinder framework. Please let me know if you need clarification on any requested changes.
All the best,
I think reviewer #2 had great suggestions for future development. I think they are beyond the scope of this specific manuscript though, with the exception of providing some example input and output files. It would be helpful for users to have these example files so they can try running the tool themselves and confirm they are getting the right output. Ideally, this would be for all common types of input formats. One simple way of doing this would be to upload some example input/output files to FigShare and then add some example usage commands using these files to the tool README.
How long does the tool take to run on typical input files and how much memory does it use? Can it be run on multiple cores, and if so, roughly what is the relative increase in run-time (e.g., is it linear)?
I think the manuscript would be greatly improved if a brief discussion was added on the following three general areas. I think a short section could be added to the discussion/results to talk about these caveats (if they are indeed caveats).
First, how sensitive are the results are to different default values for the scores (e.g., as mentioned at L320). Does tweaking these parameters typically make much difference? Also, how were the default scores chosen? It sounds like they were selected to produce reasonable results on datasets where specific systems are known to be present, but that’s not totally clear. As a potential user I would be worried that these default values would create biases when applied to genomes with different characteristics than in the training datasets. I’m guessing this kind of investigation was important when developing the first version of the tool, or for one of the other related manuscripts, but either way it would be good to quickly refer to any past validations for readers unfamiliar with the earlier work.
Second, and similarly, how confident can you be in any definitions of co-localization for specific systems across different lineages? My concern is that these co-localization relationships may be well-described in certain lineages, but that operon structure may be less in others, and so these tools might not work as well. Do you think this is a concern? And if so, what should users watch out for?
Last, how many systems are affected by the choice of whether components are allowed to overlap across systems are not (e.g., L330 and the examples later)? Knowing this would help give users an idea of whether it’s worth exploring the multi_system/multi_model options. It’s also not totally clear to me how much one might expect the overall results to change when using those options.
The tool’s link should be added https://github.com/gem-pasteur/macsyfinder to the end of the abstract
“Nanomachines” is not a commonly used term in this context (to my knowledge at least), so I recommend that the authors define it or use a simpler term
In the intro it is implied that “components” is synonymous with “proteins”. If this is the case, then I think the authors should explicitly explain why they opted for this term over simply saying proteins. I’m thinking that perhaps the authors actually mean that components can be non-protein coding genes and perhaps noncoding elements as well, which would explain why this more general term is used. Either way, this should be explained (but based on the CRISPR-Cas section I don’t think this is the case).
L33 – rather than “performing function of interest” I would reword to be “performing the same function as this system”
L34 – rather than “coherent” I would say “distinct, non-co-localized” perhaps to be clearer? (If that is what you mean there)
L67 – I think the limitations should be in a separate paragraph, as they are the key motivation for making the new version, so you want to make readers don’t miss them.
L70 – Unclear what “component-specific filtering criteria” refer to here, without looking at methods. It would be nice to see a quick example.
L77 – Instead of “novel” I think you mean “new”
L82 – should be “takes” rather than “gets”
L83 – “fasta” should be “FASTA”, upon all usages
L102 – change “have” to “had”
L120, L476 – change “…” to “, etc.”
L158: A quick description of what GA scores are based on (or an example) would be useful
E-values are well known, but I haven’t seen the term “i-evalue” before. Maybe I missed where it was defined, but if not this should be defined and distinguished from normal E-values.
L328 – should be “are” rather than “were” for both instances (i.e., use present tense when describing what the tool does in general, and not just what it did on a single occasion)
L335-336: It’s not clear to me what the edges are based on in this case (i.e., do they represent the binary relationship of whether the systems can be compatible or not? Or can they be weighted?). I think it would be useful to add a sentence clarifying that for potential users who aren’t familiar with the clique search approach.
L393 – Capitalize “archaea” and “bacteria” (here and elsewhere) to be consistent with earlier usage
L468 – rather than “we present its application” I would say “we apply it”
L499 – Make “Multi_model” lowercase (since I’m guessing the module is case sensitive?)
L524: “that the same cluster is” should be “for the same cluster to be”