Gene Expression Analysis with ArrayMining.net and GeneSpring
Disclaimer: This work has been submitted by a student. This is not an example of the work written by our professional academic writers. You can view samples of our professional work here.
Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of UK Essays.
Published: Tue, 08 May 2018
Noor Ameera Mazlan, Nurrul Shafiqah Abdullah, Siti Noorain Yousoff, Leong Wan Ting and Jasrena Rohanapi
Gene expression can be defined as a process which information of gene is used in the synthesis of a functional gene product. The analysis can differentiate between cancerous and normal tissues. A class of small non-coding RNAs which known as a microRNAs (miRNAs) control the gene expression by targeting mRNAs and trigger either translation repression or RNA degradation. Their aberrant expression may be involved in human diseases, including cancer. In addition, miRNA aberrant expression has been previously found in human breast cancer, where miRNA signatures were associated with specific clinico biological features. Here, we show that, miRNAs are also deviate expressed in breast cancer as compared to the normal breast tissue. The tools that have been chosen to identify the cancerous gene are ArrayMining and Genespring. The results produced by these tools are being compared in order to determine which results are more robust. The comparison showed that results produced from ArrayMining were more likely accurate compared to Genespring.
MicroRNAs (miRNAs) are a class of naturally occurring and small noncoding RNA. MicroRNA plays an important role in the regulation of gene expression in term of targeting mRNA and triggering RNA degradation. Mature miRNA is single stranded and having approximately 21-25 nucleotides in length. MicroRNA binds to target sites in 3’ UTR (untranslated region) of the targeted mRNA and this interaction causes mRNA degradation or block of translation.Recently, many studies show that microRNAs (miRNAs) aberrant expression will cause human disease such as cancer, neurological disease and heart disease . As a result, miRNA aberrant expression can be classified as a tumor suppression gene. In addition, miRNA are also found that aberrantly expressed in human breast cancer when compared to normal breast tissues. In this research, miRNA aberrant expression in human breast cancer is our main concern. Therefore, it is important to study the miRNA expression in between normal breast tissues and breast cancer to reveal the deregulated miRNAs in tumor tissues.
In this 21st century, with the emergence of technology in our life, there are many bioinformatics tools were developed to help scientist to do statistical and bioinformatics analysis of microRNA microarray data and study the miRNA expression profiling of normal and breast cancer tissues. So, the differences between normal and tumor breast tissues expression can be identified and improve our understanding on breast cancer disease. By this way, the deregulate miRNA being can be found out easily. Here we present five current recent use of bioinformatics tool in analyzing microarray data. The first tool is Genespring, a stand-alone software that deal with multiple array formats of data, involves multiple data display formats, consists a set of statistical clustering tools and contains automated annotation and cross-referencing . Advanced analysis tools inside GeneSpring also make itself become a very powerful microarray data analysis tool. Other than that, Genespring can classify samples into two or more by using class predictor that based on the gene expression level. Meanwhile, ArrayMining is an online web-based tool bioinformatics resource to do microarray analysis with the available features that make it stand-out . The special features provided are ensemble and consensus analysis methods, modular connections between different analysis types, new analysis approaches (e.g. BioHEL), automatic parameter selection and 2D/3D data visualization. One of the modules in this tool able to identify sets of functionally similar genes, then make summarization of gene sets into meta gene and finally apply the statistical analysis on it.On the other hand, GeneXPress is a general purpose of visualization and analysis tool that designed to support extensive post-analysis of gene expression experiments. J-Express is a Java application tools that allows analyzing gene expression which is microarray data in a way giving access to multidimensional scaling, clustering, and visualization methods in an integrated manner. The drawback of this tool is that it does not include methods for comparing two or more experiments to differentially expressed genes. Genevestigator is a microarray database which publicly available along with a statement of data analysis tools . The tool will integrates thousands of manually curated public microarray and RNAseq experiments and nice visualization of gene expression across different biological contexts will be produced (diseases, drugs, tissues, cancers, genotypes, etc.). We decided to compare the results produced by using ArrayMining and Genespring. This is because there are many algorithm provided in these tools thus the analysis produced are more robust.The sample was retrieved from Gene Expression Omnibus (GEO). The dataset was being input and raw data were normalized and ranked based on their p-value. Then it will be analyzed with different kind of algorithms.
2. MATERIAL AND METHODS
2.1 Breast Cancer Sample and Lines
Primary breast tumors from 98 samples; 34 patients who developed distant metastases within 5 years, 44 from patients that are disease-free for at least 5 years, 18 patients with BRCA1 germline mutation and 2 from BRCA2 carriers. The patients were all lymph node negative and under 55 years old while they are being diagnosed with the disease. There are about 4348 genes before being processed. 5µg total RNA was isolated from snap-frozen material and complementary RNA (cRNA)is derived by using this method. By pooling equal amounts of cRNA from each of the patients’ carcinomas, a reference cRNA pool is made.The histopathological data were associated with genes, for example, by immunohistochemical (IHC) staining (Fig. 1), oestrogen receptor (ER)-α expression can be determined. 34 tumours were clustered together in the bottom branch of the tumor dendogram for ER-α expression (ER negative).
In ArrayMining, there are 6 different modules provided for the analysis of data. For this study, we chose to use the gene selection module that applies supervised feature selection to identify differentially expressed genes. There are several algorithms that being used in this module to analyse the input data. The empirical Bayes moderated t-statistic was done for the statistical comparison that ranks genes by testing whether all pairwise contrasts between different outcome-classes are zero. Besides that, Significance Analysis in Microarrays method(SAM) is used to detect differentially expressed genes. In order to assign significance values to selected genes, this method use permutations of the measurements. PGSEA is also used to significantly identify differential expressed gene sets of functionally related genes. ENSEMBLE, an algorithm that combines the eBayes, SAM, PLS-CV and RF-MDA selection schemes to an ensemble feature ranking is also implemented in this module.
2.3 Gene Spring
Raw data were normalized and analysed using the GeneSpring software version 7.2. Expression data were median centered. By using ANOVA, statistical comparison were successfully done. It use the Benjamini and Hochberg correction for false-positive reductions. Both the Prediction Analysis of Microarray software and Support Vector Machine tool were used to determine tumors versus normal class prediction of prognostic miRNAs. Both algorithms were used for cross-validation and test-set prediction.
2.4 Analysis measure
Result will be visualized in heatmap form where it is actually a graphical representation of data where the individual values contained in a matrix are represented as colours. In GeneSpring, p-value with value p<0.05 can be classified it as cancer while ArrayMining, null hypothesis in the sample are cancer, so we identify the proportions that are classified as cancer by rejecting the null hypothesis. In this study, bad prognosis and good prognosis also can be identified by using Gene Set Analysis module in ArrayMining.
In order to show the result of bad prognosis gene and good prognosis gene in boxplot form, pre-defined cancer related gene sets obtained from R-package PGSEA is used as functional gene annotation data source. A larger number of differentially expressed gene sets were provided by PSGEA thus it will require less computation. When the result shown, it will be output of the boxplot based on four top ranked of gene samples. In these four top ranked samples, each boxplot is separate based on bad prognosis and good prognosis samples.
To identify miRNA where expression was significantly different between normal and tumor samples and could identify the different nature of these breast tissues, we used t-statistical analysis for both tools. Although the analysis used is same which is t-test, but GeneSpring show p-value result to user while ArrayMining show q-value which is the adjusted p-value found using an optimized False Discovery Rate (FDR) approach to user. GeneSpring implement ANOVA to do the t-test and show p-value while ArrayMining implement e-Bayes algorithm to do the t-test and show q-value.
3. RESULT AND DISCUSSION
There are lots of tools that can analyse gene expression in many ways. But, in our study we were only focus on tools that can produce results that coincide with our goal. ArrayMining and GeneSpring were selected, In order to compare results from these tools, same datasets have been used. For example, in this study breast cancer dataset has been used to detect which of these miRNA were diagnosing cancer. The comparative analysis of these tools will be described in details below.
Based on the results, both of these tools have heatmap analysis. Although both of these tools produce heatmap analysis, there are still differences output between them. Firstly, colour of the heatmap. In Figure 2, range colour for the heatmap is from red to yellow where red represent cancer samples while yellow represent normal samples. While range colour for ArrayMining, it is from red to green where the colour represent Z-score value. Low value of Z-score will represented in red colour and high value of Z-score represented in green colour. Obviously, that colour of heatmap in ArrayMining does not indicate whether samples are cancer or not as compared heatmap in GeneSpring. Second is from the aspect clustering. For GeneSpring software, clustering need to be done manually where the users need to click on ‘Analysis’ button first to do clustering but not in ArrayMining because clustering in ArrayMining already shown in heatmap result. Finally, is from the aspect of parameter used to produce heatmap. For GeneSpring, parameter used is p-value while for ArrayMining, parameter used is Z-score.
Figure 2: Overview of Heatmap result comparison (Left: in GeneSpring, Right: in ArrayMining)
Next, comparison that can be made is boxplot results. Boxplot allow the user to quickly see a lot of statistical information about a condition or sample. Both tools have boxplot output with different characteristics. For GeneSpring, each samples are represented in each boxplot and the samples does not separate based on bad prognosis and good prognosis just like in ArrayMining where output of the boxplot is based on four top ranked of gene samples. In these four top ranked samples, each boxplot is separate based on bad prognosis and good prognosis samples. Median, the outliers and the quantiles will be calculated between each sample in GeneSpring. Whereas, median the outliers and the quantiles will be calculated for each of prognosis in each of highest ranked genes in ArrayMining tool.
Figure 3: Overview of Boxplot result comparison (Left: in GeneSpring, Right: ArrayMining)
3.3 Determination of cancer sample
In order to validate whether the sample are cancer or not, both tools use p-value. In GeneSpring, p-value is calculated using ANOVA where genes with p-value<0.05 are classified as cancer. While in ArrayMining, to identify whether the genes are cancer or not we refer to q-values which is the adjusted p-values found using an optimised false-discovery rate (FDR) approach. By using characteristics of the p-value distribution FDR is optimised to produce a list of q-values. This approach is a more recent development where it determines adjusted p-values for each test. In our case, the null hypothesis is the samples are cancer. By using FDR approach, we identify the proportions that are classified as cancer by rejecting the null hypothesis.
3.4 Time interval used for retrieve results.
In term of time consuming used to obtain results, it found out that GeneSpring faster as compared to ArrayMining. User can get result in just a few minutes waiting. ArrayMining take longer time to display output. This is because ArrayMining is web based tool, so the speed to get the output depends on the speed of internet. User need to wait longer because the software needs to create job ID for the query request. But this not happen in GeneSpring because it is a stand-alone gene expression analysis tool, so once the user installs this tool, they can use it anytime and anywhere that they want without depend on the internet connection. Speed of both tools in displaying the results also depends on the content of database. GeneSpring has already built in database compare to ArrayMining, where Arraymining need to match results from other online databases that available.
As we can conclude, ArrayMining is the compatible tools in order to analyse and differentiate between normal and cancer tissues. This is because the analysis is more robust as the algorithm can be combined using the ENSEMBLE feature. Besides that, user can explore the tools in a shorter time and do the analysis easily thus an approximately accurate can be generated. For the future works, based on our analysis we would suggest detection of mutant cell in gene expression area.
We acknowledge support by Dr. Razib Othman Senior Lecturer of Department Software Engineering, Universiti Teknologi Malaysia, Skudai, Johor, Malaysia who supervised and guide us throughout this study by giving a lot of helpful comments.
Cite This Work
To export a reference to this article please select a referencing stye below: