Visual gene developer: a fully programmable bioinformatics software for synthetic gene optimization

(1) Gene design, analysis, and optimization

As one of the most important features, Visual Gene Developer not only contains many useful functions to facilitate gene design and analysis but also provides versatile optimization strategies. Since most basic functions are integrated into the main menu of the software, a user can easily manipulate and analyze a sequence. For example, the 'Function' section of the main menu includes parsing, translating, reversing a sequence, and generating complimentary sequence as well as silent removal of undesirable sequences and codon/mRNA optimization for a sequence in the 'Workspace' window. The 'Analysis' section of the menu covers calculating codon usage, CAI, GC content, Nc, searching multiple query sequence or repeated sequence, comparing two sequences, and predicting mRNA secondary structure. To build a codon usage table for CAI calculation or codon usage adaptation, the software has a function to directly import codon usage tables from CUTG (Codon Usage Tabulated from GenBank) website () http://www.kazusa.or.jp/codon/and calculate RSCU (Relative Synonymous Codon Usage), RSCU _max , and w value (Relative adaptiveness of a codon), and also helps users build their own local codon usage database for highly expressed genes within genomes under translational selection. Regarding sequence optimization, although a user can simply use basic functions mentioned above, the software provides a more advanced tool to solve complicated optimization problems. Using 'Gene Construct Designer' window, a user can design a gene construct that has several different gene construct components, configure optimization conditions in the 'Configuration of Gene Optimization' window, and run the optimization process in the ''Gene Optimization' window. The optimization system is based on the combination of multiple algorithms. A user can select several different algorithms to design a unique optimization strategy. For example, the simplest combination may be using 'Monte-Carlo Codon Optimization' and 'Check CAI' functions. The former function is to optimize a gene and the latter is to screen out an improper gene when its CAI value is lower than a setpoint. The implemented algorithms include performing silent removal, checking repeated sequences, restriction enzyme sites, Gibbs free energy of mRNA secondary structures, and etc.

(2) Programming capability

As already mentioned, there have been numerous trials to improve heterologous gene expression by modifying the DNA sequence, including codon optimization. However, it is still not clear what the best gene design criteria is because of insufficient knowledge of gene expression and regulation mechanisms at the molecular level, particularly for a wide variety of potential host organisms. Therefore, currently there exists no generally accepted way to optimize gene designs and many researchers are trying to develop and improve gene optimization algorithms. Of course each algorithm needs to be confirmed by experiments.

In order to provide a very flexible way to design a new algorithm, Visual Gene Developer supports script language programming and allows users to develop new functions and plug-ins to expand the functionality of the software (Figure 3). For example, although several useful gene analysis index parameters such as CAI, GC content, mRNA secondary Gibbs free energy, and Nc are known and also available in Visual Gene Developer, there are still many opportunities to identify other useful indices. To test a new finding, a user can manipulate algorithm and easily add the function to the software. To offer programming capability, the software has an integrated script engine and interface to process two popular script languages: VBScript (Visual Basic script) and JScript (Java script). A user can develop and test a new module in the 'Module Editor' window and manage modules in the 'Module Library' window to delete or edit modules. To facilitate code development, Visual Gene Developer has a unique UI called a configurable Property Bag that can contain several different properties. As a window UI object where an end-user can set parameter values, a Property Bag can be defined by the user and associated with the module. Using a specialized class, 'PropertyBag', a module developer can read or write a value of a property calling property name. In that way, a Property Bag works as a visual communication port between an end-user and a module developer to change the behavior of a module even though an end-user may not recognize user-defined algorithm and interface. Two module windows ('PropertyBag Editor' and 'PropertyBag Library') are used to define Property Bags. A user can deploy a module to a specified location of a window according to defined module category.

(3) mRNA secondary structure prediction

Visual Gene Developer carries a sophisticated mRNA prediction engine that can be used to optimize a gene construct by adjusting mRNA binding energy or structure. The engine was developed as a class by embedding the stand-alone mRNA prediction software called Vienna RNA Secondary Structure Package developed by Hofacker [28]. For module developers, Visual Gene Developer provides a user accessible class named as 'mRNApredict'. The class contains many simplified functions that can be used with a single line code in Visual Gene Developer's programming environment to predict MFE (minimum free energy) mRNA secondary structure, calculate mRNA binding energy, visualize a predicted structure, or read raw data of the predicted structure. Therefore, it is highly useful to develop more complicated optimization algorithm modules that utilize mRNA structure prediction. For example, local Gibbs free energy of mRNA structure can be repeatedly calculated to generate a binding energy profile over a test sequence.

Visual Gene Developer also includes simple menu functions and a specialized window form for mRNA structure prediction. After putting a sequence into a textbox in the 'Workspace' window or 'mRNA structure' window and clicking on menu or 'Calculation' button, a predicted mRNA secondary structure will be quickly shown in the 'mRNA structure viewer' window with a graphical representation of base pairing, probability of base pairing, energy, and locations. The 'mRNA structure viewer' window also has a function to export the image or raw data to a clipboard.

With regard to mRNA structure optimization, a common problem is that it usually requires a large number of repeated calculations of mRNA binding energy. Definitely, it takes a much longer time to find optimal mRNA structure compared with codon optimization. To reduce the overall calculation time, the mRNA prediction engine was designed and optimized for network and multi-threaded computing as will be explained later. Utilizing several computers or multiple threads in a computer, multiple processes of mRNA structure prediction can be performed at the same time to analyze multiple gene constructs or gene construct components. Thus Visual Gene Developer can be used to analyze mRNA secondary structures of several thousand genes for genome-scale studies as well as gene optimization in itself.

(4) Artificial neural network module

An artificial neural network model can be defined as a non-linear computational model that consists of a number of highly interconnected artificial neurons to simulate the structure and function of biological neural networks. The goal is mapping a set of input patterns onto a corresponding set of output patterns. It has been widely used to model complicated nonlinear systems that consist of multiple variables to predict data patterns.

Due to the potential application of artificial neural network models for gene analysis and optimization, Visual Gene Developer includes an artificial neural network prediction toolbox. The software has a configuration window to design the topology and adjust learning parameters and can directly import (or export) a data set from the clipboard or typical ASCII text format files. A typical feedforward neural network with a standard back propagation learning algorithm to train networks was implemented [29].

As with other classes, Visual Gene Developer allows a user to programmatically access the neural network toolbox and class. This means that users can utilize artificial neural network models when they develop new modules. For example, it is possible to predict gene expression levels of gene constructs in real time during the gene optimization process if a user has already developed and trained neural networks to correlate several different parameters to the expression level. In addition, the software contains useful functions to analyze the trained neural network map and test input and output variables as 2-D or ternary diagrams.

(5) Network & multi-threaded computing

Since gene analysis and optimization processes usually require extensive computations, it may take long time to find optimal genes if there are a lot of genes to analyze and/or if the gene analysis algorithms are complicated. For example, in the case of mRNA secondary structure prediction, it takes about 30 seconds to calculate a global mRNA secondary structure for a typical size gene of 1~1.5 kilo base pairs and it will take at least 3 days to generate mRNA secondary structures of 10,000 genes using a Pentium 4 computer. In order to reduce the calculation time, a grid/parallel computing using a workstation or a supercomputer is very useful [30–32]. However, the problem is that it is not easy for non-experts to access to those computing resources.

Network or multi-threaded computing can be a good alternative and Visual Gene Developer supports this approach for the calculation of mRNA structure or binding energy as well as user-defined functions. The term 'network computing system' generally refers a system that utilizes multiple computers that are connected to a network to perform computational tasks. The idea is that if several computers are available and they are connected to the internet, Visual Gene Developer connects those computers and splits the server's work load to the connected computers. For example, in order to calculate mRNA binding energy of 10 genes, Visual Gene Developer's server transfer gene sequence data to 10 client computers and then receives calculated results from them. Thus the theoretical processing time can be reduced to 1/10 of the original processing time if data transfer time is neglected. On the contrary to network computing, multi-threaded computing means a utilization of multiple threads in a single core or a multi-core computer. The concept of multiple threads or multiple processes has a close relationship with multi-tasking in a computer. For reference, a computer or an operating system runs an application on a process and each process consists of at least one thread. As an actual programming code, a thread performs a certain job. However, in spite of great advantages of multi-tasking, it significantly reduces overall performance of a single application. To overcome this intrinsic problem, Visual Gene Developer executes multiple client processes on a single computer and links them together on a local network. In this way, Visual Gene Developer simulates a multi-threaded computing system that makes use of all computer resources more efficiently.

Visual Gene Developer has a robust network & multi-threaded computing module with a user-friendly interface and provides a specialized class that contains several useful functions to monitor network connections. Because all functions including data communication protocol were implemented into a single executable file, the software doesn't require any expensive hardware and has an ability to operate both a server and multiple clients at the same time. Once computers are connected to the internet, Visual Gene Developer allows a user to easily setup a server and clients in the 'Server' window. On the client side, a user can add new clients by choosing 'Add client' in the main menu. The 'Job list' window shows the current status and includes functions to pause the service or cancel reserved jobs.

(1) CAI (Codon adaptation index)

where X _ij is the total number of the j th codon for the i th amino acid in the test gene, w _ij is the relative adaptiveness of the j th codon for the i th amino acid in the reference gene, k _i is the number of synonymous codons for the i th amino acid, and L is the total number of codons excluding AUG (Met) and UGG (Trp) in the test gene. As a special case, if w _ij is smaller than 0.01, it is adjusted to 0.01 [34].

(2) Nc (Effective number of codons)

where ${\widehat{F}}_i$ the codon homogygosity of the i th amino acid, n is the total number of the amino acid in the test gene, k _i is the total number of synonymous codons of the i th amino acid, and p _j is the codon frequency of the j th allele (synonymous codon).

where ${\bar{\widehat{F}}}_m$ (m = 2, 3, 4, or 6) is the average homogygosity for the amino acids whose total number of synonymous codons is m. For example, ${\bar{\widehat{F}}}_6=\left({\widehat{F}}_{Arg}+{\widehat{F}}_{Leu}+{\widehat{F}}_{Ser}\right)∕3$.

(3) Repeated sequence search

The software allows a user to identify repeated sequences in a test sequence. It detects not only forward directional and backward complimentary repeated sequences but also palindromic sequences and consecutively connected repeated sequences. In order to find repeated sequences, a moving window method was employed. The algorithm generates a short sequence clipped from a test sequence and then compares the partial sequence with the test sequence to find matched locations. The search process is repeated while the moving window scans along the test sequence. When the scanning is completed, potentially duplicate findings are removed if they are already included in other findings. The feature is named as the 'Smart filter' in the 'Search sequences' window.

(4) Multiple query sequence search

This function was developed to identify locations of query sequences within a sequence. A user can input a set of multiple query sequences or restriction enzyme names separated by Tab, comma (,), or Carriage Return (Enter key) in the 'Search sequences' window. For in-depth analysis a query string is split into multiple strings of single query sequence. In case of restriction enzyme names, they are converted into DNA sequences. After performing repeated searches for all query sequences in a test sequence, the software shows detailed information about the total number of occurrences and their locations in a gene for every query sequence. A user can choose one of a predefined sequence set such as common restriction enzyme sites, potential intron cryptic splice sites or polyadenylation signal sequences.

(5) Profile calculation of CAI, mRNA Gibbs free energy, or GC content

The software contains 3 implemented modules that are used to calculate a profile of CAI, mRNA binding energy, or GC content of a test sequence. Their algorithms are quite similar between them as the moving window approach was equally adopted and their codes were developed from the same template code. In general, any single calculation such as GC content can be repeatedly performed while a moving window is sliding over a test sequence. The procedure is initiated when a moving window is located at the first base of the test sequence. For example, mRNA binding energy of the first 60 bases is calculated if the size of the moving window is set to be 60 bases that can be adjusted by the user. After the first calculation, the moving window steps forward to the next location such as to the 11th base of the test sequence if the step size of the moving window is 10 bases. In this way, the RNA binding energy is repeatedly computed at an interval of 10 bases until the moving window arrives at the end of the sequence. To generate data for a profile plot, both the location of the moving window and its corresponding mRNA binding energy are recorded as a table format. Since the codes were written in VBScript, a user can easily modify source codes to develop new profiling functions.

(1) Codon optimization

The software provides a predefined module that is based on a well-known Monte-Carlo simulation or a predefined probability table [13, 15, 19, 20]. It utilizes a codon usage table and replaces original codons with new ones while maintaining the identity of the same amino acids. To be specific, Visual Gene Developer not only has a function to import codon usage tables from CUTG (Codon Usage Tabulated from GenBank) but also provides a manual edit mode for the target codon usage map and allows a user to generate a local database of reference sets of optimal codon usage tables. The software automatically calculates the RSCU, RSCU _max , and w _i values, and then generates a look-up table (LUT) of synonymous codons. For example, if alanine has four synonymous codons such as GCA, GCC, GCG, and GCT whose expected fractions are 0.1, 0.2, 0.3, and 0.4, respectively, the LUT will consist of 100 GCAs from 1 to 100, 200 GCCs from 201 to 300, 300 GCGs from 301 to 600, and 400 GCTs from 601 to 1000 in a memory array. Finally, one of 1000 codons is randomly chosen and then it replaces the original codon. By utilizing the look-up table, it is possible to perform codon optimization very quickly. In addition, the software has a pre-defined function that allows a user to keep track of changes in codon usage bias as a graphical representation. Meanwhile, since the current version of the software doesn't have a built-in database of optimal codon usage maps of highly expressed genes, a user may need to rely on other available sources including papers and web databases where a user can get an optimal codon usage map for a specific host genome and then put it into Visual Gene Developer.

(2) mRNA optimization

In order to optimize a gene in terms of mRNA binding energy, the algorithm was developed utilizing both mRNA prediction and codon optimization modules. At the code level, an original sequence is continuously modified until its Gibbs free energy is in a specific range given by minimum or maximum Gibbs free energy where the modification refers to synonymous substitution. With regard to the modification strategy, the simplest approach is that the number of mutations is gradually increased one by one when the calculated Gibbs free energy is out of range. Meanwhile, the base module was also used to develop more complicated modules. For example, a binding energy profile of a long test sequence can be optimized by repeatedly applying the base algorithm to all local sequences with a moving window method. In this way, all local mRNA structures can be optimized while minimizing the number of base changes. Similarly, a user can increase or decrease binding energy at specific locations in a sequence. Visual Gene Developer has a specialized window for mRNA optimization for a typical user and provides related class functions for a module developer.

(3) Removal of undesirable sequences

The coding region of a gene may include undesirable sequences such as restriction enzyme sites, potential polyadenylation signal sequences, or cryptic splice sites. Visual Gene Developer provides a function to remove such unwanted sequences without changing the resulting amino acid sequence. The algorithm is based on the synonymous substitution and similar to that for codon optimization except it replaces only a few codons with correspondent synonymous codons in the target sequence region that needs to be modified (Figure 4). The first step is to identify the location of the target DNA sequence in a gene and then determine the location of the site for synonymous substitution. To simplify the substitution process, terminal sequences of the target sequence are truncated if they are located outside of complete codons. The current version of the software carries 4 relevant modules that are written in VBScript. A user can easily remove undesirable sequences including predefined potential polyadenylation signals and intron cryptic splice sequences.

(1) Excluding query sequences or specified restriction enzyme sites

The purpose of the module is to avoid undesirable DNA sequences including potential cryptic splice sites, polyadenylation signal, and restriction enzyme sites. When those sequences are found, the gene construct will be excluded from the candidate gene construct list.

(2) Checking stability of mRNA secondary structure

This algorithm is a modification of the 'mRNA Gibbs free energy plot'. It is used to analyze mRNA secondary structures of all partial sequences of a sequence. If the calculated Gibbs free energy of a local sequence in a moving window is lower than a threshold value, the module returns 'Not pass' value and consequently the gene construct will be screened out.

(3) Removing repeated sequences

In order to prevent repeated sequences in a gene construct, the module is developed to count the total number of repeated sequences in a sequence. If the number is more than a prescribed set point, the gene construct will be ruled out.