Gene clustering tool

The authors observed that approximately a third of all metabolic genes in A. Two algorithms that rely on information based on Gene Ontology GO or gene expression data are designed to predict all gene clusters from a query genome sequence and are not necessarily restricted to finding only metabolic gene clusters. We include these methods here as their algorithmic logic can be repurposed in future iterations of metabolic gene cluster prediction tools and for prioritizing functional metabolic gene clusters.

C-Hunter developed by Yi et al. The algorithm uses GO terms in a directed acyclic graph DAG highlighting increasingly specific functional categories. C-Hunter builds clusters by searching for shared GO terms, while keeping track of the order of genes on a chromosome. The algorithm detected 54, 25 and clusters in Escherichia coli , Saccharomyces cerevisiae and A. By comparing predictions with known E. Furthermore, C-Hunter was able to accurately predict 4 of 6 and 3 of 3 genes in S.

In another study, Wada et al. Furthermore, they deduced 34 clusters consisting of anywhere between 3 and 22 genes as being statistically significant at a false discovery rate of 0. These 34 clusters contained genes corresponding to a variety of cellular processes including cell cycle, metabolism, signal transduction, transcription and transport. A total of 12 clusters contained at least one gene that was annotated for metabolism and 9 clusters contained at least two metabolic genes.

Four clusters deemed to be statistically significant contained a majority of metabolic genes [ 33 ]. Although not strictly used as methods for detection of novel gene clusters, these tools can be convenient in the analysis for substrate specificities of previously predicted clusters and complementary to NP.

Unlike NP. An updated version of the tool, NRPSpredictor2 [34], can process both bacterial and fungal sequences.

SBSPKS allows for assessing substrate specificity of query catalytic domains by comparing active site motifs with those contained in domains of known specificity. Growing evidence indicates that metabolic gene clusters are not limited to microbial genomes [ 14 ]. To build new or extend existing algorithms to find metabolic gene clusters from more complex genomes such as animal and plant genomes, key challenges related to genome complexity, genome quality, stringency criteria for signature enzymes and tandem duplicates, as well as availability of experimental data should be considered.

Plant and animal genomes are larger in size than microbial genomes with a substantial amount of noncoding sequences that can lie in between genes [ 47 ]. Consequently, model parameters for physical distance trained on bacteria may not be suitable for gene cluster prediction in higher organisms. Bacterial genomes tend to be compact; with longer and more complex gene structures in eukaryotic genomes [ 47 ], the likelihood for uncertainty is increased, which can directly affect definitions of cluster boundaries.

In addition, gene clusters may be sparse, i. An adaptive approach is more desirable: one that allows for clusters where genes belonging to a particular pathway may either be tightly packed or spread out. Algorithmically, one potential adaptive approach would be to judge the cluster boundaries based on gene co-expression data.

Enzymatic genes that are co-expressed with a cluster core can be grouped together as one cluster as opposed to excluding them if they may not be present within a fixed cluster length away from the core region. Moreover, existing methods have not accounted for splicing variants of genes.

There is evidence for splicing variants of regulators such as NAC TFs in Populus trichocarpa demonstrating different functional activity [ 48 ]. However, there is overall limited knowledge about how splicing variants impact the function of enzymes and in particular substrate specificity. The power of gene cluster prediction depends on the quality of the genome assembly and annotation, given that for many existing algorithms the location of signature enzymes and the presence of adjacent tailoring enzymes are vital attributes.

A widely used metric to assess genome quality is a contig N50 [ 49 ]. Scaffold N50 is computed similarly using scaffolds instead of contigs.

Genomes with small N50s are not appropriate for cluster prediction. In addition, it is also important to assess completeness in terms of gene content. Performing a BUSCO analysis on an animal or plant genome of interest will be important prior to undertaking any kind of gene cluster prediction routine. The logic behind many existing algorithms follows two key assumptions: 1 if a signature enzyme is identified, then this is used as the seed for a gene cluster, and 2 all detected gene clusters will encode a multi-step metabolic pathway geared toward the synthesis of a specialized metabolite [ 14 ].

However, not all signature enzymes are expected to occur in gene clusters. Algorithms need to differentiate between such nonclustered signature enzymes and those that should be included as part of gene clusters. Furthermore, predicted gene clusters may contain tandemly duplicated genes that do not necessarily encode for multiple enzymatic steps [ 14 ]. Chae et al. Incorporating a stringent rule-set in the prediction algorithm such as the need for multiple contiguously located enzymes that together catalyze more than one reaction could help increase accuracy of predictions.

For many animal and plant species, there remains the necessity of generating additional experimental data for purposes of a prioritization, and b validation of gene cluster predictions. Gene expression data can be used to improve the accuracy of gene cluster predictions.

Gene co-expression analyses played an important role in revealing that the tomatine and solanine gene clusters in tomato and potato, respectively, were represented as two distinct clusters on separate chromosomes in each species [ 51 ]. Condition-specific co-expression data can be informative in highlighting whether predicted gene clusters are active only during particular experimental conditions.

In addition, protein—protein interaction data, which can provide information on interactions between metabolic enzymes and potential pathway structure, can also be useful [ 52—54 ]. Ultimately, incorporation of these types of experimental data could help not only improve the accuracy of cluster prediction, but also prioritize predicted clusters for experimental validation. For validating gene cluster predictions, methods have relied on altering the expression of clustered genes in a native or heterologous context.

In a native context, a few example strategies include a experimental knock-outs, and b experimental knock-ins to replace the promoters of clustered genes that may otherwise be silent under particular conditions [ 7 ]. When coupled with an assessment of changes in metabolite profiles, these strategies have led to the detection of novel compounds [ 7 ]. Gene clusters may also be expressed in genetically or molecularly tractable hosts such as S.

Additionally, the use of various computational methods can provide results that complement each other or that may be conflicting. Validation of gene cluster predictions with experimental data can aid in improving the accuracy of and resolving any conflicts between the bioinformatics tools. High-quality genome sequencing combined with the development of novel bioinformatics pipelines is enabling systematic identification of metabolic gene clusters and the discovery of novel specialized metabolites.

In this review, we provide an overview of computational methodologies, concepts and caveats of metabolic gene cluster prediction. The methods developed so far have been largely focused on clusters in bacteria and fungi and key challenges and opportunities exist for extending these tools or developing new tools for the prediction of clusters in more complex genomes.

We recently became aware of several new gene cluster prediction methods devised for plant genomes see pre-print articles [ 55 ],[ 56 ],[ 57 ]. In addition, a new algorithm building upon the work by Chae et al. These tools should help accelerate the elucidation of plant specialized metabolic pathways.

As the biosynthetic pathways for many plant specialized metabolites are unknown, identification and validation of plant metabolic gene clusters presents an exciting opportunity for discovery of a new metabolic pathways, b mechanisms of chemical diversity in plants and c a treasure trove of specialized metabolites aimed at fighting human disease.

Key Points The discovery of novel metabolic pathways is being enabled by the increasing availability of high-quality genome sequencing coupled with the development of bioinformatics toolkits to identify metabolic gene clusters. Additionally, we compare and contrast key aspects of their algorithmic logic. Bioinformatics methods have to date largely focused on identifying metabolic gene clusters in bacteria and fungi. The complexity of many animal and plant genomes necessitates new innovation toolkits to be developed to identify metabolic gene clusters from these genomes.

Arvind Chavali is a postdoctoral associate at the Carnegie Institution for Science. He received his PhD in bioengineering from the University of Virginia. She received her PhD in biological sciences from Stanford University.

Davies J. Specialized microbial metabolites: functions and origins. J Antibiot Tokyo ; 66 7 : — 4. Google Scholar. Towards a new science of secondary metabolism. J Antibiot Tokyo ; 66 7 : — Role of secondary metabolites in defense mechanisms of plants. Biol Med ; 3 2 : — The role of flavonoids in the establishment of plant roots endosymbioses with arbuscular mycorrhiza fungi, rhizobia and Frankia bacteria. Plant Signal Behav ; 7 6 : — Biomed Res Int ; : Callaway E , Cyranoski D. Anti-parasite drugs sweep Nobel Prize in medicine Nat News ; : —5.

Recent advances in natural product discovery. Curr Opin Biotechnol ; 30 : — 7. From hormones to secondary metabolism: the emergence of metabolic gene clusters in plants. Formation of plant metabolic gene clusters within dynamic chromosomal regions. Delineation of metabolic gene clusters in plant genomes by chromatin signatures. Nucleic Acids Res ; 44 5 : — Osbourn A. Secondary metabolic gene clusters: evolutionary toolkits for chemical innovation. Trends Genet ; 26 10 : — Plant metabolic clusters - from genetics to genomics.

New Phytol ; 3 : — Medema MH , Osbourn A. Computational genomic identification and functional reconstitution of plant natural product biosynthetic pathways.

Nat Prod Rep ; 33 8 : — Nutzmann HW , Osbourn A. Gene clustering in plant specialized metabolism. Curr Opin Biotechnol ; 26 : 91 — 9. Computational approaches to natural product discovery. Nat Chem Biol ; 11 9 : — Bioinformatics approaches and software for detection of secondary metabolic gene clusters. Methods Mol Biol ; : 23 — Automated genome mining for natural products. BMC Bioinformatics ; 10 : ClustScan: an integrated program package for the semi-automatic annotation of modular biosynthetic gene clusters and in silico prediction of novel chemical structures.

Nucleic Acids Res ; 36 21 : — CLUSEAN: a computer-based framework for the automated analysis of bacterial secondary metabolite biosynthetic gene clusters. J Biotechnol ; : 13 — 7. Nucleic Acids Res ; 43 W1 : W — SMURF: genomic mapping of fungal secondary metabolite clusters. Fungal Genet Biol ; 47 9 : — Bioinformatics ; 32 8 : — Accurate prediction of secondary metabolite gene clusters in filamentous fungi.

FunGeneClusterS: predicting fungal gene clusters from genome and transcriptome data. Synth Syst Biotechnol ; 1 2 : — 9. Investigation of terpene diversification across multiple sequenced plant genomes. Insights into secondary metabolism from a global analysis of prokaryotic biosynthetic gene clusters. Cell ; 2 : — Phylogenomic analysis of natural products biosynthetic gene clusters allows discovery of arseno-organic metabolites in model Streptomycetes.

Genome Biol Evol ; 8 6 : — Motif-independent prediction of a secondary metabolism gene cluster using comparative genomics: application to sequenced genomes of Aspergillus and ten other filamentous fungal species. DNA Res ; 21 4 : — MIDDAS-M: motif-independent de novo detection of secondary metabolite gene clusters through the integration of genome sequencing and transcriptome data.

PLoS One ; 8 12 : e Genomic signatures of specialized metabolism in plants. You can adjust them according to needs. We use Clinker for MGCs alignment and visualization. Note that the -p parameter is used by default to generate an interactive HTML web page where you can modify the MGCs figure and export it as a publication-quality file. This is a simple example to help you quickly start MagCluster journey.

We use the genomes of Candidatus Magnetominusculus xianensis strain HCH-1 see the paper and Magnetofaba australis IT-1 see the paper to show how it works. You can see the video tutorial here or follow the detailed tutorial below.

We recommand the evalue to be set as 1e Note that the --outdir , --prefix , --locustag and --compliant parameters are used by default. The reference MGCs file that we provide is also used with --proteins. Now that we have HCH Good job!

Now you can open the. Use Clinker to generate a MGCs figure. We recommand to use the -o parameter to generate a MGCs alignment file where you can browse the homologous gene similarities among genomes. Be Careful! If there is complete genome s in your dataset, the alignment process will take unreasonable time.

In that case, we recommand you skip the alignment process with -na no alignment parameter. Now you should have an interactive html opened in your browser. You can adjust the figure as you like and export it as a. Skip to content. Filter annotations: Minimum number of levels for annotations:. Maximum number of levels for annotations:. Delimiter: Comma. Number of column annotations:. Number of row annotations:. Quotes: single or double single double no quotes. Missing values:.

Gene subsetting or clustering: Subset a pathway. Subset a custom gene list. Cluster genes. Choose one cluster. Paste your gene list below: Clear data Genes should be separated by white space spaces, tabs or newlines. Number of k-means clusters:. Cluster ID to choose:. Column annotation groups to keep:. Column annotations to keep:. Collapse columns with similar annotations: no collapse median mean. Maximum percentage of NAs allowed in rows:.

Maximum percentage of NAs allowed in columns:. Row scaling: no scaling unit variance scaling Pareto scaling vector scaling. PCA options change data options.

Data options Principal component on x-axis: 1. Principal component on y-axis: 1. Display options change coloring options. Color grouping:. Confidence level for ellipses:. Line width for ellipses:. Line type for ellipses: solid dashed dotted dotdash longdash twodash. Shape grouping:. Shape scheme: letters various. Plot width:. Margin ratio:. Point size:. Legend position: none right bottom left top.

Plot labels Font size:. Prefix for axes' labels:. Plot options Plot type: Points only. Violin plot with points. Box plot with points. Violin plot only. Box plot only. Separate overlapping points: no. Jittering width:. Heatmap options change data options. Data options show imputed values. Clustering distance for rows: no clustering correlation Euclidean maximum Manhattan Canberra binary. Clustering method for rows: single complete average McQuitty median centroid Ward Ward unsquared distances.

Tree ordering for rows: tightest cluster first higher median value first higher mean value first lower median value first lower mean value first original reverse original. Number of clusters in rows:. Clustering distance for columns: no clustering correlation Euclidean maximum Manhattan Canberra binary. Clustering method for columns: single complete average McQuitty median centroid Ward Ward unsquared distances. Tree ordering for columns: tightest cluster first higher median value first higher mean value first lower median value first lower mean value first original reverse original.

Number of clusters in columns:. Display options Column annotations:. Row annotations:. Heatmap color range maximum:. Heatmap color range minimum:. Cell border: no border grey60 grey black. Plot labels General font size:. Font size of numbers:. Precision of numbers:.

Font size of row names:. Font size of column names:. Export options ID for settings:. Metsalu, Tauno and Vilo, Jaak. Clustvis: a web tool for visualizing clustering of multivariate data using Principal Component Analysis and heatmap.