FN-Identify: Novel Restriction Enzymes-Based Method for Bacterial Identification in Absence of Genome Sequencing

Sequencing and restriction analysis of genes like 16S rRNA and HSP60 are intensively used for molecular identification in the microbial communities. With aid of the rapid progress in bioinformatics, genome sequencing became the method of choice for bacterial identification. However, the genome sequencing technology is still out of reach in the developing countries. In this paper, we propose FN-Identify, a sequencing-free method for bacterial identification. FN-Identify exploits the gene sequences data available in GenBank and other databases and the two algorithms that we developed, CreateScheme and GeneIdentify, to create a restriction enzyme-based identification scheme. FN-Identify was tested using three different and diverse bacterial populations (members of Lactobacillus, Pseudomonas, and Mycobacterium groups) in an in silico analysis using restriction enzymes and sequences of 16S rRNA gene. The analysis of the restriction maps of the members of three groups using the fragment numbers information only or along with fragments sizes successfully identified all of the members of the three groups using a minimum of four and maximum of eight restriction enzymes. Our results demonstrate the utility and accuracy of FN-Identify method and its two algorithms as an alternative method that uses the standard microbiology laboratories techniques when the genome sequencing is not available.


Introduction
Bacterial identification is an important routine in the clinical and industrial microbiology laboratories. Microbiologists and researchers stepped up their efforts to improve and facilitate the rapid characterization of various microbial communities. Traditional bacterial identification strategies are mainly based on morphological, biochemical, enzymatic, antigenic, staining, and antibiogram characterization [1]. However, these strategies are time consuming and sometimes fail to identify the bacteria accurately [2]. Many other strategies appear to have improved bacterial identification accuracy, such as automated cellular fatty acid (CFA) analysis, yet these strategies require expensive system and standardized culture condition. Moreover, it cannot differentiate closely related species such as Escherichia coli and Shigella [2]. Protein analysis and phage analysis are also used as methods for bacterial identification [3]. With the presentation and rapid progress of molecular biology and molecular markers, several new and enhanced bacterial identification methods were developed. These methods include plasmid analysis [4], restriction fragment length polymorphism (RFLP) [5], pulsefield gel electrophoresis (PFGL) [6], random amplified polymorphism DNA (RAPD) [7], fluorescent in situ hybridization (FISH) [8], and DNA Props [9].
In the early 1980s, polymerase chain reaction (PCR) provided novel approaches for bacterial identification through amplification of specific sequences/genes from the bacterial genome. Several ribosomal RNA (rRNA) genes and Internal Transcribed Spacers (ITSs) had been utilized for PCR-based 2 Advances in Bioinformatics bacterial identification such as 16S rRNA, 23S rRNA, 5S rRNA, and SSU rRNA [8,10]. The PCR-based identification uses the ribosomal genes, since ribosomal genes play an important role in living organisms and have functional stability over evolution ages due to rare variation in its sequences through millions of years, which makes them suitable to be used for identification and taxonomical purposes.
Numerous ribosomal RNA genes and ITSs such as Hsp65, rpoB, gyrB, groEL, and recA have been tested as a genetic marker in bacterial identification [11]. However, 16S rRNA is the most widely used ribosomal RNA genes in bacterial identification due to several reasons: (1) the 16S rRNA gene presents in almost all bacterial families; (2) it has functional and evaluation stability; (3) in many cases, multiple copies of the 16S rRNA gene presented in the genome and sometimes differences in sequences present as well, which can be used to distinguish closely related species; (4) the sequence length is about 1500 : 1550 bp, which is enough for taxonomical purpose and suitable for amplification; (5) the 16S rRNA gene sequence contains conserved regions and variable regions; therefore, it is possible to design a universal primer on these conserved regions for gene sequence amplification [1,12]. Therefore, several methods for 16S rRNA amplification and analysis were developed: ribotyping [13], denaturing gradient gel electrophoresis (DGGE) [14,15], temperature gradient gel electrophoresis (TGGE) [15], amplified ribosomal DNA restriction analysis (ARDRA) [16], and terminal restriction fragment length polymorphism (T-RFLP) [17,18].
With the rapid progress in DNA and RNA sequencing technology, sequencing of 16S rRNA gene and several other genes became a popular method for bacterial identification and phylogenetic reconstruction. Furthermore, it is employed in nucleic acid-based detection, quantification of microbial diversity, and discovery of novel bacterial isolates in different microbiology laboratories [19][20][21][22].
Despite the outstanding advancements in speed and accuracy and the remarkable decrease in cost of the sequencing technologies in the recent years, sequencing technologies in developing countries are out of reach for the majority of clinical and research laboratories. This is mainly due to the high cost of establishing sequencing facility and high cost of reagents and maintenance [23][24][25]. Furthermore, the lack of trained personnel and the limited access to up-to-date scientific information play an important role in constraining the use of such indispensable technology in many clinical and industrial microbiology laboratories in these countries [26,27]. Most labs depend on outsourcing the DNA/RNA sequencing through using commercial services. Typically, the sample is prepared and sent to a local company that sends it to companies in the EU or China to be sequenced and the results are sent back. Based on our observations, this process is expensive and time consuming (up to several months) and can fail at any point.
In this work, we present a FN-Identify, an efficient and sequencing-free bacterial identification method, as a proposed alternative that can be employed when genome sequencing is inaccessible. FN-Identify, which stands for fragment number-identify, is based on techniques that are available in most of the standard microbiological laboratories.  Our new method depends on sequences available in Gen-Bank and other public databases, such as RDP-II [28], Silva [29], and Greengenes [30], restriction enzymes, and the two FN-Identify algorithms that we developed ( Figure 1). We used bacterial population of 33 members (species and strains) of Lactobacillus genus to develop the method and used two other bacterial populations of 33 and 22 members (species and strains) of Pseudomonas and Mycobacterium, respectively, to test the method. FN-Identify successfully identified and differentiated all the species/strains using two different genes 16S rRNA and HSP60, in two independent analyses. The identification scheme and the utilized restriction enzymes, created by FN-Identify, demonstrate its efficiency as a rapid, accurate, and affordable alternative method for bacterial identification in the absence of the sequencing technologies.  had to annotate the 16S rRNA sequences, as its annotation was unavailable in the database. We picked up the 16S rRNA sequences from L. kefiranofaciens ZW3 genome using the same primers successfully used with all other Lactobacillus members. The two primers picked up four copies of 16S rRNA sequences ( Table 2 strain ID 28).

16S
rRNA Primer Selection. We tested 13 different primer sequences obtained from 8 published studies (Table 3). We used Python script to test the primers and compare the sequence positions we got using each primer with 16S rRNA position in the genome annotation in (NCBI), to confirm that the primer would pick the 16S rRNA sequence. Based on this testing, we selected two primers (Table 3, 8F and 1541R) from [31]. The two selected primers appear in all Lactobacillus genomes in this study and with the largest product length (1550 pb).

HSP60 Primer Design.
A universal degenerate primer for picking up HSP60 sequences was designed based on the conserved regions in the HSP60 extracted sequences. We identified the conserved regions by performing multiple sequence alignment (MSA) using CLC Sequence Viewer software (CLC Bio, Swansea, UK). Table 3 shows the sequences of the designed and forward and reverse primers.

Restriction Enzymes and Restriction Map.
We collected the information about restriction enzymes and restriction sites from the database of restriction enzymes (REBASE), Roberts 1980 andRoberts et al., 2010 [39, 40], and the restriction enzyme database attached to the DNA Star software (DNASTAR Inc., Madison, WI, USA). Prediction of the in silico restriction map was performed using the restriction sites information and the seqBuilder tool of Lasergene software tool (DNASTAR Inc., Madison, WI, USA).
Advances in Bioinformatics 5 Our Annotation for 16S rRNA sequences in L. kefiranofaciens ZW3.

Genomics in the Developing Countries.
Currently, genome sequencing is the technology-of-choice for several research and clinical applications due to its rapid development, remarkable speed, continuously improved accuracy, and affordable sample processing cost. However, in several developing countries, the genome sequencing technologies are still out of reach for most of researchers and scientists due to several reasons which constrain employing such indispensable technology. Firstly, the high cost of establishing sequencing facility and high cost maintaining the facility in poorresources countries. Secondly, the lack of well-trained personnel to run the facility. Thirdly, the weak power, Internet, and computational infrastructures. Finally, the limited access to the updated scientific data, literature, and training [26,27]. The scientific community expected this problem over a decade ago with the rising of the next-generation sequencing technologies [25]. In the following years, many developing countries took steps to utilize these technologies by establishing institutions for genomics and provide funds to facilitate running and maintaining them as well as hiring and training personnel. Reports about case studies in several developing countries including Mexico, Thailand, South Africa, and India show the efforts made to import these technologies and the expected impact on research, public health, and economic development in these countries [41]. Despite these improvements, the problem seems to be still far from being solved, especially in Africa [23,26], letting the researchers with one choice, that is outsourcing. This situation raises the need of developing alternative methods that can be utilized in doing standard research tasks until the availability of sequencing technologies. 6 Advances in Bioinformatics

Obtaining Standard Dataset of Bacterial Genomes and
Genes. The identification of the family of certain bacteria is usually based on the morphological and other characteristics of the colony, while the identification of the species and strains requires molecular and more sophisticated methods [2,16,42,43]. Therefore, we selected the Lactobacillaceae family as a representative of bacterial population with several industrial and health importance [44][45][46][47] to be used in developing FN-Identify method and algorithms. In addition, Lactobacillus members have different important genes used in bacterial identification and barcoding such as 16S rRNA and HSP60 with several differences in sequences and copy numbers. This makes Lactobacillus members ideal for developing and testing new methods for bacterial identification based on the analysis of the restriction patterns of its genes. We downloaded the 33 complete Lactobacillus genome sequences and annotations available in the NCBI (Table 1). According to the genome annotations, all Lactobacillus genomes have one copy from HSP60 and between four and nine copies of 16S rRNA, except for Lactobacillus kefiranofaciens ZW3 (strain ID: 28, Table 1), where its genome annotation shows absence of 16S rRNA ( Table 2). For Lactobacillus kefiranofaciens ZW3 we annotated the 16S rRNA gene using the selected 16S rRNA universal primers (see below). At least two of 16S rRNA copies are in the reverse direction. Strains under the same species have the same number of 16S rRNA copies except Lactobacillus johnsonii strains (strain IDs: 17 and 19, Table 1) since one of them has four copies and the other has six. Tables 1 and 2 list Lactobacillus species/strains used in this study as well as the copy numbers, start and end of each copy, and an ID that we gave to each species/strain that we will use hereafter.

Primer Selection and Design.
In order to select standard universal primer(s) for 16S rRNA sequences from all Lactobacillus genomes, we tested several primers from publish literature (Table 3). We performed the in silico screening for each primer using the separated gene sequences as well as the whole genome sequences. Our primers in silico screening show that (8F) and (1541R) primers present in most of the separated 16S rRNA gene sequences with largest product length (see Table 3 for primer sequences). Therefore, we keep the sequences between both primers and exclude all other sequences, including the primers sequences.
In some cases, these two primers are not present in 16S rRNA separated sequences. For instance, the two primers failed with the separated 16S rRNA genes of the strain Lactobacillus salivarius UCC118 (strain ID: 32, Table 1). However, when we used them with the whole genome of the same strain we found 8F and 1541R beginning from nucleotides 74,520 and 76,053, in agreement with the genome annotation of the first 16S rRNA copy (from 74,540 to 76,056). Similarly, Lactobacillus salivarius CECT 5713 (strain ID: 31, Table 1) has the same difference.
In some cases, there was a difference in length between the 16S rRNA returned in silico sequence and the length of the 16S rRNA in the genome annotations. For instance, Lactobacillus johnsonii (strain IDs: 17 and 19, Table 1) returned a 1555 bp sequence when using the two selected primers, while the gene length in the genome annotation was 1650 bp. However, it is within the start and end of the annotated gene, so we accept it. Apart from these few cases, the selected 16S rRNA primers 8F and 1541R performed perfectly with all Lactobacillus genomes. This guarantees that the returned in silico sequences will agree with the isolated sequences in lab.
After selecting the 8F and 1541R primers as universal primers for 16S rRNA, we used them to annotate the 16S rRNA gene in the Lactobacillus kefiranofaciens ZW3 (strain ID: 28, For HSP60 gene, we could not find a universal primer in the published literature. Therefore, we design a universal primer based on the conserved nucleotide sequences of HSP60. The conserved nucleotide sequences were identified be multiple sequence alignment (MSA) using CLC Sequence Viewer software (CLC Bio, Swansea, UK). Based on the alignment results, we were able to design two degenerate primers for HSP60 (HSP60-F and HSP60-R, Table 3): the forward primer (HSP60-F) 5 ATGGCWAARGANNTHAARTT3 and the reverse primer (HSP60-R) 5 TCDGCVACNACN-GCTTCNGA3 yielded in 1560 bp for all species while the annotated HSP60 is 1600 bp. Again, we take the sequences between both primers and exclude all other sequences, including the primers sequences.

In Silico Restriction Map.
In order to perform an in silico enzymatic restriction for the 16S rRNA and HSP60 genes, we selected 12 commercially available restriction enzymes from hundred of enzymes that we collected their data. To select these 12 enzymes, we scanned all enzymes using Python script and the information of the restriction site that we collected from the database of restriction enzymes (REBASE) [40], the restriction enzyme database attached to the DNA Star software (DNASTAR Inc., Madison, WI, USA), and other resources [39], against the 16S rRNA and HSP60 sequences. The selected enzymes have different restriction sites, which will help us differentiate the Lactobacillus species through the differences in restriction maps of the selected gene sequences. Next, we performed an in silico enzymatic restriction for the 16S rRNA and HSP60 gene sequences using seqBuilder tool of Lasergene software tool (DNASTAR Inc., Madison, WI, USA).
The in silico enzymatic digest results in DNA fragment lengths ranges approximately from 10 bp to 1570 bp. Since the very short fragments are unobservable in the experiments, we excluded the fragments length less than 30 bp [48]. Although it is expected that the number of return DNA fragments = the number of restriction sites + 1, the results are different from the expected ones and this is mainly due to two reasons: firstly some fragments being equal in length or the difference in lengths being too small to be observed in the gel separation and secondly our exclusion of the very short fragments.
The exclusion of the short fragments was observed in several species and strains from those we used in this study. For instance, Lactobacillus delbrueckii subsp. bulgaricus 2038 (strain ID: 9, Table 1) has six restriction sites for HinfI enzyme but the number of the return DNA fragments was four only. This is because one of the fragments was of length 9 bp, two other fragments are with length of 119, and two other fragments are with very close length (difference is less than 10 bp) [49]. The same strain has five restriction sites for TfiI but the return DNA fragments contain one fragment of length 9 bp. Therefore, it returns five fragments only. Table S5 contains the details of the return DNA fragments for each restriction enzyme.
Other sources of differences in ribotyping between the Lactobacillus genomes are the variation in the 16S rRNA copy numbers between different species and the differences in sequences between the multiple copies within the same genome (Table 2). This leads to difference in restriction sites and number of restriction fragments. One noticeable example for this phenomenon is the Lactobacillus brevis ATCC 367 (strain ID: 5, Table 1), which contains five copies of 16S rRNA genes with three different sequences ( Table 2). The restriction of these three different sequences with HinfI enzyme results in four, five, and six DNA fragments since they have three, four, and five HinfI restriction sites, respectively (Table S5). The same three different sequences of 16S rRNA contain two, three, and three restriction sites for TfiI enzyme, respectively. Another example is Lactobacillus fermentum (strain IDs: 12 and 13) that shows similar behavior with the HinfI enzyme ( Table 2 and Table S5) To determine the number of returned DNA fragment from a particular species/strain that contains several copies of 16S rRNA sequences, we compare the lengths of the fragments and exclude the duplicated equal fragments length. This is how the restriction will be done actually in the lab, as the fragments with the same length will be in the same band in the gel. For instance, for Lactobacillus brevis ATCC 367 (strain ID: 5) there are five different copies of 16S rRNA with three different sequences (see above) ( Table 2). Restriction with HinfI enzyme returned five fragments for two copies and four for the other copy. After excluding the duplicated fragment lengths, we have seven fragments only in the gel (976 bp, 891 bp, 379 bp, 243 bp, 136 bp, 117 bp, and 86 bp). Supplementary Figure 1 shows comparison of two cases where he fragments number is equal to the expected and where it is not.
For HSP60 gene, the construction of the restriction map was straightforward. Each Lactobacillus species or strain contains one single copy of the gene (Table 2). Therefore, the differentiation between them will be based on differences in restriction patterns between species/strains (Table S7).

FN-Identify Method
This section describes our proposed sequencing-free bacterial identification method in detail. The proposed method identifies bacterial species/strains based on the number of fragments and/or fragment lengths that result from the restriction of certain genes using a given set of restriction enzymes. Therefore, we refer to it as the fragment numberidentification method or FN-Identify. The main goal of FN-Identify is to establish an identification scheme for bacterial species utilizing fragments patterns of enzymatic restrictions such as the restriction map we built in the above section. The established scheme specifies the set of enzymes that could be employed to identify a given (unknown) gene sequence as well as the order of their application. The identified gene refers to a particular species/strain within the restriction map.
The idea behind FN-Identify is inspired from two basic observations. First, the number of fragments resulting from each restriction of a DNA sequence (e.g., 16S rRNA gene sequence) would differ based on the employed restriction enzyme. Generally speaking, a given gene sequence G could be split into and fragments if two different enzymes, and , were employed, respectively, where and are likely to be different. Second, some restriction enzymes are more discriminative than other enzymes with respect to different bacterial families. Assume that both and are employed to cut all sequences belonging to a specific bacterial family F. Let be a set containing the number of fragments resulting from cutting all sequences of F using . Enzyme is said to be more discriminative than if and only if | | > | |, where | | denotes the cardinality of set .
For the purpose of illustration, consider an extreme example where all sequences of F are split into the same number of fragments if is employed, that is, | | = 1, whereas each sequence of F produces a different number of sequences if is employed; that is, | | = , where is the number of sequences of F. Clearly, while can identify all sequences of F perfectly, does not provide any useful information for discriminating the sequences of F. FN-Identify benefits from the above two observations to create an identification scheme for bacterial genes utilizing only a set of discriminating restriction enzymes. The proposed method consists of two algorithms. The first algorithm, CreateScheme, aims at finding an efficient identification scheme given a bacterial family F and the adopted set of restriction enzymes E. The second algorithm, GeneIdentify, employs the obtained scheme to identify a given unknown gene sequence.
The CreateScheme algorithm (see Algorithm 1) recursively builds a tree that represents an identification scheme for species/strains of F. Each node of consists of three components, namely, the processed group of species/strains, the restriction enzyme that produces the largest number of distinct number of fragments when applied to that group, and the number of distinct number of fragments produced. Obviously, the first component of the root node of ( Figure 2) consists of all species/strains of F and the third component should be 1 since all species/strains of F consist of only one fragment, that is, the whole sequence. Once the enzyme that produces the largest number of distinct number of fragments, for all members of F, is found, it should be assigned to the second component of root( ). Algorithm 1 can be described informally as follows.
Step 1. Predict the restriction map RM by restricting all species/strains of F using all enzymes of E.
Step 2. Search RM to find the restriction enzyme F max that gives the largest number of distinct number of fragments when applied to all the gene sequences of F.
Step 3. Use results obtained from the application of F max to assemble species/strains of F into different groups according to the resulting number of fragments such that strain sequences that are split into the same number of fragments are grouped together in the same category. As an example, Figure 2 shows that the restriction-enzyme F max categorizes the species/strains of F into three different groups, namely, 1 1 , 1 2 , and 1 3 , where the superscript indicates the tree level (level 1). All species/strains in these groups are fragmented, using F max , into 3, 4, and 7 fragments, respectively.
Step 4. Apply Step 3 recursively to each resulting group consisting of more than one species/strains. For example, the illustrative example in Figure 2 shows that the first group in level 1, 1 1 , is then split into two different groups, 2 1 and 2 2 in level 2, where species/strains in these groups are fragmented into two and three fragments employing the restriction enzyme Input: Bacterial gene G and a gene-identification scheme Output: identified-gene or failure currentNode ← root( ) repeat Cut G using currentNode.enzyme Chld ← currentNode.getChildren fragList ← number of fragments of all nodes in ℎ if G.numOfFragments ∉ fragList then return failure else find a child ∈Chld that has number of fragments equal to G.numOfFragments and save it as currentNode.chld; currentNode ← currentNode.chld end if until a leaf node is met return identified-gene

Strain #3
· · · · · · Figure 2: An example of a tree representing an identification scheme. Dotted lines points to a strain that is identified.
Step 5. The algorithm stops if either (1) the number of species/strains of all groups being processed is one or (2) no further application of any restriction enzyme can discriminate species/strains in groups containing more than one species/strains. The former case indicates that the algorithm can identify all species/strains of F using the adopted set of restriction enzymes E. The second case, on the other hand, takes place if some species/strains cannot be identified employing E. In this case, another factor, such as the fragment length, can be utilized to break any potential ties among unidentified species/strains.
Once an identification scheme is created for F, it would be possible to identify an unknown gene sequence G as belonging to F or not by traversing starting from the root node following the GeneIdentify algorithm (see Algorithm 2). The first step is to visit the root node of to specify the restriction enzyme that should be employed first to cut G, that is, F max . Then, the number of fragments of all children (groups) of the current node (root) is retrieved and compared to the number of fragments resulting from cutting G using F max . The node of the matched group is then visited and its associated restriction enzyme is retrieved and applied to G in order to decide which node has to be visited in the next level, and so on. This process is continued until a leaf node is met. If such a node is found, the processed gene sequence will be successfully identified as the species/strains at that (leaf) node. Otherwise, the identification process fails. As mentioned earlier, if there are no matching groups at any level of , a different factor such as lengths of fragments could be tried and the identification process will continue.
The GeneIdentify algorithm can be illustrated further using the example shown in Figure 2. Let a strain G be one of the strains, referred to as strain #3, that belongs to F. In this example (see dashed lines), G is identified by applying the following sequence of restriction enzymes: F max , max . This is because G is split into three fragments if F max is employed and two fragments if 1 1 max is employed and no other species/strain is fragmented into the same number of fragments if 2 1 max is employed to cut F.

Developing FN-Identify Method.
In order to develop our proposed method and algorithms, we used the 16S rRNA sequences of a population of 33 members of Lactobacillus (Table 1), an example of bacteria with genes with multiple copies in the genome ( Table 2). FN-Identify and the two algorithms were able to identify and differentiate between the 33 species/strains based on the fragment numbers of the 16S rRNA sequences using six restriction enzymes (Figure 3, Supplementary Table 5). For a given species/strains a minimum of one enzyme and maximum of five enzymes were required for the identification (Figure 3 strains ID: 5 and 8, resp.). By adding the fragment length as a second factor, FN-Identify successfully identify and differentiate between the 33 species/strains using five restriction enzymes only. Furthermore, a maximum of three enzymes only was required for the identification of any given species/strains (Figure 4, Supplementary Table 6).
To further improve the identification efficiency of FN-Identify method and algorithms, we used the HSP60 genes as an example for genes with a single copy in the genome ( Table 2). Genes represented with a single copy provide less Enzyme name Fragments Species/strain ID (Table 1)  variations in sequences (see above). Therefore FN-Identify might require more restriction enzymes to differentiate the 33 species/strains or even may fail in identifying some of them. However, with further tuning of the algorithms, FN-Identify shows comparable performance to what it does in the genes represented with a multiple copies (16S rRNA). It was able to identify the 33 species/strains based on the fragment numbers using six restriction enzymes (Supplementary Figure 2 and Supplementary Table 7). For a given species/strains a minimum of two enzymes and maximum of five enzymes were required for the identification (Supplementary Figure  2 strains ID: 24 and 33, resp.). When we used the fragment length as a second factor, FN-Identify required four restriction enzymes only to identify the 33 species/strains (Supplementary Figure 3 and Supplementary Table 8). Moreover, a maximum of three enzymes only was required for the identification of a given species/strains (Table 4, Supplementary Figure 3, and Supplementary We obtained the sequences of the 16S rRNA genes of 22 members of Mycobacterium and 33 members of Pseudomonas using the same approach that we used with Lactobacillus (See Section 2). The variations in the 16S rRNA copy number and differences sequences between the multiple copies within the same genome appear in Pseudomonas, whereas the Mycobacterium genomes of the 22 members contain only one or two 16S rRNA copies (Supplementary Tables 2 and  4). We applied FN-Identify on the two testing datasets and FN-Identify successfully identified all the members of the two groups using the fragments numbers only and eight and seven enzymes to identify the 33 members of Pseudomonas and the 22 members of Mycobacterium, respectively (Table 4). Furthermore, for a given species, a maximum of eight and seven enzymes and minimum of seven and five enzymes were required to identify a given member of the Pseudomonas and Mycobacterium groups, respectively (Table 4, Supplementary  Figures 4 and 6, and Supplementary Tables 9 and 11). By adding the fragment length as a second factor, FN-Identify successfully identifies the species/strains of the two groups using seven and four restriction enzymes for Pseudomonas and Mycobacterium groups, respectively. Furthermore, a maximum of seven and four enzymes and a minimum of four and three enzymes were required for the identification of any given species/strains (Table 4, Supplementary Figures 5 and  7, and Supplementary Tables 10 and 12).
Collectively, these results demonstrate the efficiency and utility of the FN-Identify method and the two developed algorithms in identifying bacterial species/strains within a genus and show that the method is applicable in bacterial groups with distinct properties.  2 Members with differences in 16S rRNA sequences. In some cases two or more members have 100% similarity in 16S rRNA sequences. Those members are considered as one entry to FN-Identify. 3 The maximum and minimum number of enzymes required identifying a given member of the group.

Advances in Bioinformatics
Enzyme name Fragments Species/strain ID (Table 1) Fragment number and sizes

Applications and Future
Perspective. The assessment of FN-Identify method and the two developed algorithms shows the potentials of the method, with standard microbiology protocols and instruments. FN-Identify is a computational method that is designed as an aid that helps designing and minimizing the experimental procedures required for bacterial identification. Ideally, FN-Identify interfaces with the experimental and clinical workflows through receiving inputs (expected bacterial group, gene(s) to be used for identification, and list of restriction enzymes) and provides outputs that lead the later bench exterminates (list and order of enzymatic restriction experiments and the identification scheme that is used to interpret the experimental results).
To be fully utilized, FN-Identify needs a software tool that is connected with a database of gene sequence (e.g., 16S rRNA and HSP60) in different bacterial families and database of restriction enzymes. The software should implement the two algorithms and automate the selection of the species and the enzymes as well as automating building the restriction map and the identifying scheme. We are currently building this tool as a webserver that provides these services for free to enable the scientific community in the developing countries to utilize FN-Identify.

Conclusion
Bacterial identification is an important routine that is required in several microbiological and environmental applications and research. The current techniques are highly dependent on genome sequencing techniques that target certain genes that present almost in all bacterial species. Although the genome sequencing techniques observed outstanding improvements in accuracy and decrease in cost, developing countries remain far from employing these indispensable technologies due to several barriers. Therefore, alternative sequencing-independent methods are required to facilitate the needed tasks with affordable costs and using the available facilities. We developed FN-Identify method, a sequencing-independent method for bacterial identification, using standard microbiological protocols and instruments, restriction enzymes, and two algorithms that we developed (CreateScheme and GeneIdentify). FN-Identify was tested against standard bacterial populations of 22 and 33 bacterial species/strains of the Mycobacterium and Pseudomonas groups, respectively. The method successfully differentiate all the species/strains in two independent analyses using two different genes 16S rRNA and HSP60 for each of the two groups. A webserver is being developed for FN-Identify to automate the scheme building and maximize the utilization of the method. We believe that FN-Identify is a useful alternative to the sequencing methods when they are out of reach.