A Complete Gene Expression Analysis from Raw Data to Biological Insights Using Matlab

 Author : Waqas Javaid

Abstract

This study demonstrates a comprehensive computational pipeline for analyzing high-dimensional gene expression data, from raw measurements to biological interpretation. Using synthetic transcriptomic data simulating two experimental groups, we perform essential preprocessing including log-transformation and variance filtering to enhance biological signals [1]. Principal Component Analysis reveals underlying sample structures and group separations, while hierarchical and k-means clustering identify sample relationships [2]. A core differential expression analysis, employing statistical testing with false discovery rate correction, successfully identifies genes significantly altered between conditions, visualized through volcano plots. The workflow culminates in a heatmap displaying expression patterns of significant genes across samples [3]. This end-to-end analysis showcases a standardized, reproducible approach to transform raw expression matrices into actionable biological insights, highlighting key genes and patterns that distinguish experimental conditions [4]. The methods are broadly applicable to omics data, providing a foundational framework for exploratory bioinformatics.

1. Introduction

The rapid advancement of high-throughput technologies has generated vast amounts of gene expression data, presenting both a remarkable opportunity and a significant challenge for modern biology.

This data, which captures the dynamic activity of thousands of genes simultaneously, holds the key to understanding fundamental biological processes, disease mechanisms, and potential therapeutic targets [5]. However, the sheer volume and complexity of this information render traditional analytical methods insufficient, necessitating robust computational pipelines to extract meaningful biological signals from inherent noise [6]. A typical analysis must navigate several critical stages: rigorous preprocessing to normalize and clean the data, dimensionality reduction to reveal underlying structures, and sophisticated statistical modeling to identify genes with biologically relevant changes in expression.

Table 1: Dataset Parameters

This article presents a complete, step-by-step workflow for analyzing gene expression datasets, implemented in the accessible MATLAB environment [7]. We begin by generating a synthetic dataset that mirrors real-world experimental conditions, featuring controlled differential expression between two sample groups. The pipeline then guides the reader through essential steps including log-transformation, variance-based filtering, and exploratory data analysis via clustering and Principal Component Analysis (PCA) [8]. The core of the investigation focuses on differential expression analysis to pinpoint genes significantly altered between conditions, with results visualized through intuitive plots like volcano charts and expression heatmaps [9]. This integrated approach provides a clear, reproducible framework that transforms raw numerical data into interpretable biological insights, empowering researchers to validate hypotheses and discover novel patterns within their own transcriptomic studies [10].

1.1 The Data Revolution and Its Challenge

The genomic revolution, powered by technologies like microarrays and RNA sequencing, has fundamentally transformed biological research. We can now measure the expression levels of tens of thousands of genes across hundreds of samples in a single experiment, creating a comprehensive snapshot of cellular activity [11]. This wealth of data promises unprecedented insights into development, disease progression, and cellular responses to stimuli. However, this opportunity comes with a formidable computational challenge. The resulting datasets are high-dimensional, noisy, and complex, often described as having a “large p, small n” problem many more genes (variables) than samples (observations). This complexity obscures the true biological signals we seek, making raw data virtually uninterpretable without sophisticated analytical tools [12]. The primary goal shifts from data generation to data interpretation, requiring a bridge between massive numerical matrices and actionable biological knowledge.

1.2 The Need for a Structured Analytical Pipeline

To navigate this complexity, researchers require a systematic, reproducible pipeline a standardized sequence of computational steps that methodically processes raw data into reliable results. An effective pipeline must address several core tasks: ensuring data quality and comparability through normalization, reducing dimensionality to visualize and understand sample relationships, and applying rigorous statistics to identify key genes driving observed differences [13]. Without such a structure, analyses become ad-hoc, prone to bias, and difficult to reproduce or validate. This structured approach is not just a technical formality; it is the foundation of robust, credible scientific discovery in computational biology [14]. It transforms a daunting data deluge into a logical flow, where each step builds upon the verified output of the last, gradually refining the signal and mitigating noise.

1.3 From Raw Numbers to Biological Readiness (Preprocessing)

The journey begins with preprocessing, a critical phase that prepares raw expression values for downstream analysis. Raw data is often skewed by technical artifacts, varying scales, and outliers that can distort all subsequent findings.

Table 2: Sample Group Distribution

Common first steps include a log2 transformation, which stabilizes variance across the wide dynamic range of expression data and makes its distribution more symmetrical [15]. Following this, normalization techniques, such as Z-scoring, adjust the data so that each gene has a mean of zero and a standard deviation of one, ensuring fair comparison across all measurements. A crucial filtering step often removes genes with very low or invariant expression, as they contribute little information and can increase noise. This preparatory work is essential; it turns the raw, messy data into a clean, standardized matrix ready for exploration and statistical testing.

1.4 Visualizing Structure and Relationship (Exploratory Analysis)

With clean data in hand, exploratory data analysis (EDA) seeks to uncover the overarching structure and relationships within the samples. Techniques like Principal Component Analysis (PCA) are indispensable here. PCA reduces the thousands of gene dimensions into a few key components that capture the most variance in the data, allowing us to plot samples in 2D or 3D space. This visualization can reveal whether samples cluster by their known experimental groups, hint at outliers, or suggest unexpected batch effects [16]. Complementing this, hierarchical clustering builds a tree (dendrogram) based on expression similarity, providing another view of sample-to-sample relationships. This stage answers the initial, vital question: Does the global structure of my data reflect the experimental design, and are there any major anomalies?

1.5 The Core Question: Finding Differential Expression

The central objective for many gene expression studies is differential expression (DE) analysis.

Table 3: Top Differentially Expressed Genes

This step aims to identify which specific genes show statistically significant expression changes between predefined conditions such as healthy versus diseased tissue or treated versus control samples. This is typically done using statistical tests (e.g., t-tests) that compare the expression distribution of each gene across groups. However, testing thousands of genes simultaneously creates a multiple testing problem, dramatically increasing the chance of false positives [17]. Therefore, raw p-values must be adjusted using methods like the False Discovery Rate (FDR) to control for these errors. The output is a curated list of genes ranked by both their statistical significance and the magnitude of their change (fold-change), providing a high-confidence target list for biological interpretation.

1.6 Interpretation, Visualization, and Biological Insight

The final stage translates statistical results into biological understanding. Powerful visualizations are key. A volcano plot displays the entire DE results, plotting statistical significance against the magnitude of change, making it easy to spot the most promising genes [18]. Heatmaps then show the expression patterns of these significant genes across all samples, visually confirming their group-specific behavior and revealing potential subpatterns or co-regulated gene groups. The ultimate goal is to take this refined list of genes and connect it back to biology using enrichment analysis to see if they belong to common pathways, consulting databases for known functions, and forming new, testable hypotheses. This closes the loop, turning abstract numerical output into a coherent biological story about the system under study.

1.7 Validating and Contextualizing Findings

Once a list of differentially expressed genes (DEGs) is established, the critical task of biological validation and interpretation begins. The identified genes, while statistically significant, require contextualization within known biological frameworks. This involves functional enrichment analysis using tools and databases like Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), or Reactome to determine if the DEGs are overrepresented in specific biological processes, molecular functions, or cellular pathways [19]. This step moves the analysis from a simple list of gene identifiers to a mechanistic understanding for instance, revealing that the upregulated genes are predominantly involved in “inflammatory response” or “cell cycle progression.” Simultaneously, validation using independent datasets or experimental methods such as qRT-PCR is crucial to confirm the computational findings, ensuring the results are not artifacts of the specific analytical pipeline or dataset.

1.8 Advanced Modeling and Network Analysis

To move beyond individual genes and understand the system’s dynamics, advanced modeling techniques are employed. This includes constructing gene co-expression networks, where genes with highly correlated expression patterns across samples are linked, potentially indicating functional relationships or common regulation [20]. Methods like Weighted Gene Co-expression Network Analysis (WGCNA) can identify modules (clusters) of tightly co-expressed genes and correlate these modules with sample traits. Furthermore, if prior knowledge is available, one can perform pathway analysis or even attempt to reconstruct regulatory networks to infer causality, identifying potential transcription factors or signaling hubs that drive the observed expression changes. This systems-level perspective is essential for understanding the complex, interconnected nature of cellular processes rather than viewing genes in isolation.

1.9 Assessing Robustness and Reproducibility

A rigorous analysis must include an assessment of its own robustness and the reproducibility of its results. This involves sensitivity analyses, such as repeating the differential expression analysis with different normalization methods, variance filters, or statistical models to see if the core set of significant genes remains stable [21]. Techniques like cross-validation can be applied in a classification context to assess how well the expression signature predicts sample groups. Additionally, the entire computational workflow should be documented and shared using tools like R Markdown, Jupyter Notebooks, or version-controlled scripts (e.g., on GitHub). This ensures the analysis is transparent and reproducible by other researchers, a fundamental pillar of credible computational science that allows findings to be independently verified and built upon.

1.10 Translation and Application

The ultimate step translates the computational insights into potential applications. In biomedical research, this means connecting the molecular signature to clinical outcomes: Can the gene expression profile serve as a diagnostic biomarker, predict patient prognosis, or indicate therapeutic response? This might involve survival analysis if clinical data is linked to the samples. In agricultural or biotechnological contexts, the findings could guide genetic engineering or breeding programs. The final output is no longer just a report of statistical results, but a data-driven narrative with clear, actionable implications [22]. This step answers the “so what?” question, defining the practical impact of the analysis and proposing concrete next steps for experimental validation or clinical investigation to move the discovery from in silico to in vitro or in vivo.

2. Problem Statement

The central challenge in modern transcriptomics lies not in data generation, but in the effective distillation of high-dimensional, noisy gene expression matrices into reliable biological insights. Researchers are frequently confronted with datasets where the number of measured genes vastly exceeds the number of biological samples, creating a statistical and interpretative bottleneck. Without a systematic and reproducible analytical framework, it is exceptionally difficult to separate true biological signals from technical noise, validate findings against false discoveries, and visualize complex relationships within the data. Consequently, there is a pressing need for a clear, step-by-step computational pipeline that standardizes the journey from raw expression values to a curated list of differentially expressed genes and their biological context. This workflow must integrate rigorous preprocessing, robust statistical testing, and intuitive visualization to empower biologists who may lack deep computational expertise to confidently interpret their results, form testable hypotheses, and ensure their analyses are both transparent and reproducible.

You can download the Project files here: Download files now. (You must be logged in).

3. Mathematical Approach

The mathematical foundation of this pipeline employs linear algebra for dimensionality reduction via Principal Component Analysis (PCA), which decomposes the data covariance matrix to identify orthogonal axes of maximum variance. Statistical inference utilizes the Student’s t-test to calculate p-values for differential expression between sample groups, followed by probabilistic correction for multiple hypothesis testing using the False Discovery Rate (FDR). Distance metrics, such as correlation-based measures, underpin hierarchical clustering, while centroid-based optimization defines the k-means algorithm. Transformations, including the logarithmic and Z-score functions, are applied to stabilize variance and normalize expression distributions, respectively, ensuring data conforms to the assumptions of downstream parametric analyses. The pipeline’s mathematical core begins with data standardization via log-transformation and Z-scoring to meet statistical assumptions. Differential expression analysis then employs a two-sample t-test for each gene, with resulting p-values adjusted using the Benjamini-Hochberg procedure to control the False Discovery Rate (FDR) and account for multiple comparisons.

Principal Component Analysis performs dimensionality reduction by solving the eigendecomposition of the data covariance matrix, projecting samples onto orthogonal axes of maximal variance.

Sample clustering utilizes a correlation-based distance metric for hierarchical linkage and minimizes the within-cluster sum of squares in k-means.

Finally, results are synthesized in a volcano plot, visualizing the relationship between the statistical significance and biological effect size for all genes.

The mathematical operations begin by transforming the raw data using a base-two logarithm, which compresses its wide dynamic range and makes the variance more stable across all expression levels. Following this, each gene’s expression profile is standardized by subtracting its mean and dividing by its standard deviation, a process known as Z-scoring, which centers the data and gives every gene an equal weight for comparison. To identify genes that differ between groups, a standard t-test is calculated for each gene, comparing the means of the two sample groups. Because thousands of such tests are performed simultaneously, the resulting p-values are adjusted using a procedure that controls the False Discovery Rate, estimating the expected proportion of false positives among the list of declared significant genes. For visualizing sample relationships, Principal Component Analysis computes new axes called principal components by finding the directions in the data that capture the most variance, which involves calculating and breaking down a covariance matrix into its eigenvectors and eigenvalues. In parallel, clustering employs a correlation-based distance measure to quantify similarity between samples, using an average linkage method to build a hierarchical tree, while the k-means algorithm iteratively assigns samples to clusters by minimizing the total squared distance from each point to its assigned cluster center.

4. Methodology

The methodology for this gene expression analysis follows a structured, eight-stage pipeline designed to transform raw data into interpretable biological insights. First, synthetic gene expression data is generated to simulate two experimental conditions, creating a controlled dataset with known differentially expressed genes for method validation. The raw expression matrix then undergoes critical preprocessing, beginning with a log-two transformation to stabilize variance across its wide dynamic range, followed by Z-score normalization to standardize each gene’s profile to a mean of zero and unit variance. A variance filter is applied to remove genes with low expression variability, retaining only the most informative features for downstream analysis and reducing noise [23]. The core analytical phase employs Principal Component Analysis on the transposed, filtered matrix to reduce dimensionality and visualize sample relationships in a lower-dimensional space defined by orthogonal axes of maximum variance. Sample clustering is performed using two complementary methods: hierarchical clustering with average linkage based on a correlation distance matrix, and partition-based k-means clustering applied to the principal component scores. Differential expression analysis constitutes the statistical heart of the workflow, where a two-sample t-test compares expression levels between groups for each gene, with resulting p-values adjusted for multiple testing using the Benjamini-Hochberg false discovery rate correction. Finally, the results are synthesized and visualized through multiple layers of interpretation [24]. A volcano plot illustrates the relationship between statistical significance and biological effect size for all genes, while a targeted heatmap displays expression patterns of the most significant genes across all samples. The complete workflow, from synthetic data generation through statistical testing to final visualization, is implemented in a single, reproducible MATLAB script, ensuring transparency and facilitating validation of each analytical step [25].

5. Design Matlab Simulation and Analysis

The simulation generates a controlled, synthetic gene expression dataset designed to mimic a real-world experiment for methodological validation. We create a matrix representing one thousand genes measured across sixty biological samples, evenly split into two distinct experimental groups. The baseline expression for all genes is modeled as normally distributed values centered around a typical mean, simulating the background transcriptional noise present in any biological system. To introduce a known biological signal, one hundred and twenty genes are randomly selected to be artificially “differentially expressed.” For these target genes in the second sample group, we systematically add a substantial positive shift to their expression levels, combining a fixed increase with additional random variation. This process simulates a true, biologically relevant upregulation event. To complete the dataset, systematic identifiers are generated for each gene and sample, providing clear labels for all downstream analysis and visualization steps. This approach results in a data structure with a known ground truth we precisely know which genes are altered and between which groups allowing us to rigorously test the entire analytical pipeline’s ability to correctly recover these simulated signals from within the background noise.

This figure presents the initial distribution of the simulated data before any processing. The raw expression values, generated from a normal distribution with an added positive shift for a subset of genes, typically show a right-skewed profile. This skewness is common in real transcriptomic data, where a few highly expressed genes can create a long tail. The histogram’s shape provides a visual baseline for understanding the data’s original scale and variability. Observing this raw distribution is a critical first step, as it justifies the need for subsequent transformation. The goal of later normalization steps is to reshape this distribution to meet the assumptions of standard statistical tests, which this initial plot helps to benchmark.

This figure shows the powerful effect of data preprocessing. The log2 transformation compresses the wide dynamic range of the raw data, mitigating the right-skew observed in Figure 1 and stabilizing the variance across different expression levels. Following this, row-wise Z-score normalization centers each gene’s expression profile around zero with a standard deviation of one. The resulting distribution, as seen in the histogram, becomes approximately symmetric and bell-shaped, resembling a standard normal distribution. This transformation is essential because it makes expression levels comparable across different genes and conditions, a fundamental requirement for valid downstream statistical comparisons and clustering analyses.

This Mean-Variance plot is a key diagnostic tool in high-throughput biology. It visualizes a common property of expression data where the variance of a gene’s measurements often depends on its mean expression level. In many datasets, highly expressed genes show greater variability. The plot helps identify this trend and informs the filtering strategy. A horizontal threshold line, typically based on a percentile of the variance (e.g., the 50th percentile), can be drawn to select genes with the most variable expression across samples. Filtering based on this metric removes genes with little to no informative change, thereby reducing noise and computational load while focusing the analysis on the most dynamically regulated features.

You can download the Project files here: Download files now. (You must be logged in).

The scree plot is used to determine the intrinsic dimensionality of the data. It plots the cumulative variance captured by each principal component (PC) in descending order. A steep initial slope indicates that the first few PCs explain a large proportion of the total data variability, suggesting strong underlying patterns or batch effects. The point where the curve begins to plateau (the “elbow”) often guides the decision on how many PCs to retain for further analysis, such as visualization or clustering. This figure demonstrates that much of the structure in the filtered gene expression data can be summarized in a low-dimensional space, justifying the use of PCA for effective data reduction and exploration.

This is the primary visualization for assessing global sample relationships. Each point represents one sample projected onto the first two principal components, which are the directions of greatest variance in the data. Samples are colored according to their pre-defined experimental group. The plot reveals whether the major sources of variation in the data correlate with the biological question in this case, group separation. Clear clustering of samples by their true labels suggests that the gene expression differences between groups are strong enough to be captured by the dominant patterns in the data. This provides an initial, unsupervised confirmation that the experimental groups are transcriptionally distinct.

The dendrogram presents an alternative, tree-like visualization of sample similarities. The vertical axis represents the distance (or dissimilarity) between samples or clusters, while the leaves at the bottom are the individual samples. The structure shows how samples are sequentially merged based on the similarity of their expression profiles. We expect samples from the same biological group to merge at shorter distances, forming distinct branches on the tree. This unsupervised clustering method allows us to see if the samples naturally group according to the experimental design without prior knowledge of the labels, serving as a check on the PCA results and revealing any potential sub-structure or outliers within the groups.

This figure overlays the results of a partition-based clustering algorithm (k-means) onto the PCA visualization. Unlike hierarchical clustering, k-means requires pre-specifying the number of clusters (here, k=3). The plot shows how the algorithm partitions the two-dimensional PC space into distinct groups. Comparing these machine-assigned clusters to the true labels (Figure 5) evaluates the algorithm’s performance. Discrepancies can reveal interesting insights: for example, a k-means cluster might split a true biological group, suggesting internal heterogeneity, or it might combine parts of both groups, indicating that the major principal components do not perfectly align with the experimental factor of interest.

You can download the Project files here: Download files now. (You must be logged in).

Following statistical testing, this plot provides a focused, detailed view of a gene identified as highly significant. The boxplot displays the median, interquartile range, and spread of expression values for that specific gene in Group 1 versus Group 2. It offers unambiguous visual evidence of the differential expression detected by the t-test. The clear separation between the boxes and their whiskers confirms the gene’s strong association with the experimental condition. This type of plot is crucial for validating individual hits from a high-throughput analysis, moving from a list of statistical scores to a concrete visual example of the biological effect.

The volcano plot is a fundamental tool for summarizing the results of a genome-wide differential expression analysis. Each point represents one gene. The x-axis shows the log2 fold change (biological effect size), and the y-axis shows the negative log10 of the adjusted p-value (statistical significance). Genes with large magnitude changes appear far to the left or right, while highly significant genes appear high on the plot. The most biologically interesting genes are typically found in the top-left or top-right corners. Threshold lines (e.g., |logFC| > 1 and FDR < 0.05) can be drawn to define significance cutoffs, with points passing both thresholds (often highlighted in red) representing the final list of high-confidence differentially expressed genes for further study.

The heatmap provides a global overview of the expression patterns for all significant genes across all samples. Rows represent genes, columns represent samples (often grouped by condition), and color intensity represents expression level (from low in one color to high in another). This visualization serves multiple purposes: it confirms that significant genes show coherent, group-specific patterns (e.g., a block of red in Group 2 samples), it can reveal sub-clusters of genes with similar behavior (potentially co-regulated genes), and it acts as a final, integrated summary of the analysis outcome. It translates the numerical results from statistical tests and lists into an intuitive visual format that encapsulates the core transcriptional differences between the experimental conditions.

6. Results and Discussion

The analysis pipeline successfully identified clear transcriptional differences between the two simulated experimental groups. Principal Component Analysis revealed distinct clustering of samples based on their true group labels along the first principal component, indicating that the major source of variance in the dataset corresponds to the defined biological condition. Hierarchical clustering further supported this separation, with dendrogram branches largely grouping samples by their experimental origin [26]. Differential expression analysis, employing a t-test with stringent false discovery rate correction, yielded a robust list of statistically significant genes, a subset of which was prominently visualized in a volcano plot. This plot effectively showcased genes with both large effect sizes and high statistical confidence, confirming the pipeline’s ability to recover the known simulated signal from background noise. The heatmap of significant genes provided a compelling visual summary, displaying coherent, group-specific expression patterns that reinforce the biological validity of the findings. These results demonstrate the efficacy of a structured computational workflow in transforming raw data into interpretable biological insights. The clear separation in PCA and clustering validates the experimental design and the strength of the simulated differential expression [27]. The statistical rigor applied through variance filtering and FDR control ensures that the reported gene list minimizes false positives, a critical consideration in high-dimensional biology. The complementary visualizations from the global overview of the volcano plot to the detailed patterns in the heatmap serve to communicate complex results accessibly [28]. This end-to-end approach, from normalization and dimensionality reduction to statistical testing and visualization, provides a reproducible template for analyzing gene expression data, effectively bridging raw numerical output and meaningful biological discovery.

7. Conclusion

In summary, this study demonstrates a complete and reproducible computational pipeline for extracting meaningful biological signals from high-dimensional gene expression data. By implementing a sequential workflow of preprocessing, dimensionality reduction, clustering, and rigorous statistical testing on a synthetic dataset, we validated the method’s ability to accurately identify differentially expressed genes [29]. The clear separation of groups in PCA and clustering plots, coupled with the statistically robust gene list from the volcano plot and the coherent patterns in the heatmap, confirms the pipeline’s efficacy [30]. This structured approach provides a critical framework for transforming complex, noisy transcriptomic data into reliable, interpretable results, thereby empowering researchers to generate and validate biological hypotheses with greater confidence and reproducibility.

8. References

[1] M. Schena, D. Shalon, R. W. Davis, and P. O. Brown, “Quantitative monitoring of gene expression patterns with a complementary DNA microarray,” Science, vol. 270, no. 5235, pp. 467–470, 1995.

[2] M. B. Eisen, P. T. Spellman, P. O. Brown, and D. Botstein, “Cluster analysis and display of genome-wide expression patterns,” Proceedings of the National Academy of Sciences of the USA, vol. 95, no. 25, pp. 14863–14868, 1998.

[3] B. M. Bolstad, R. A. Irizarry, M. Åstrand, and T. P. Speed, “A comparison of normalization methods for high density oligonucleotide array data,” Bioinformatics, vol. 19, no. 2, pp. 185–193, 2003.

[4] G. K. Smyth, “Linear models and empirical Bayes methods for assessing differential expression,” Statistical Applications in Genetics and Molecular Biology, vol. 3, no. 1, 2004.

[5] J. Quackenbush, “Computational analysis of microarray data,” Nature Genetics, vol. 32, pp. 496–501, 2002.

[6] R. Gentleman, V. Carey, D. Bates, B. Bolstad, and others, “Bioconductor: Open software development for computational biology,” Genome Biology, vol. 5, no. 10, 2004.

[7] O. Alter, P. O. Brown, and D. Botstein, “Singular value decomposition for genome-wide expression data processing,” Proceedings of the National Academy of Sciences of the USA, vol. 97, no. 18, pp. 10101–10106, 2000.

[8] M. Ringnér, “What is principal component analysis?” Nature Biotechnology, vol. 26, no. 3, pp. 303–304, 2008.

[9] I. T. Jolliffe, Principal Component Analysis, 2nd ed., Springer, 2002.

[10] W. E. Johnson, C. Li, and A. Rabinovic, “Adjusting batch effects in microarray expression data,” Biostatistics, vol. 8, no. 1, pp. 118–127, 2007.

[11] Y. Benjamini and Y. Hochberg, “Controlling the false discovery rate,” Journal of the Royal Statistical Society, Series B, vol. 57, no. 1, pp. 289–300, 1995.

[12] J. D. Storey and R. Tibshirani, “Statistical significance for genomewide studies,” Proceedings of the National Academy of Sciences of the USA, vol. 100, no. 16, pp. 9440–9445, 2003.

[13] O. Troyanskaya, M. Cantor, G. Sherlock, P. Brown, and others, “Missing value estimation methods for DNA microarrays,” Bioinformatics, vol. 17, no. 6, pp. 520–525, 2001.

[14] D. M. Witten and R. Tibshirani, “A framework for feature selection in clustering,” Statistical Science, vol. 25, no. 4, pp. 456–468, 2010.

[15] T. Hastie, R. Tibshirani, and J. Friedman, The Elements of Statistical Learning, 2nd ed., Springer, 2009.

[16] R. Xu and D. Wunsch, “Survey of clustering algorithms,” IEEE Transactions on Neural Networks, vol. 16, no. 3, pp. 645–678, 2005.

[17] J. MacQueen, “Some methods for classification and analysis of multivariate observations,” Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, 1967.

[18] L. Kaufman and P. J. Rousseeuw, Finding Groups in Data: An Introduction to Cluster Analysis, Wiley, 1990.

[19] J. A. Hartigan and M. A. Wong, “A k-means clustering algorithm,” Applied Statistics, vol. 28, no. 1, pp. 100–108, 1979.

[20] M. B. Eisen, “Clustering analysis of gene expression data,” Methods in Enzymology, vol. 303, pp. 179–205, 1998.

[21] M. E. Ritchie, B. Phipson, D. Wu, Y. Hu, C. W. Law, W. Shi, and G. K. Smyth, “limma powers differential expression analyses,” Nucleic Acids Research, vol. 43, no. 7, 2015.

[22] M. I. Love, W. Huber, and S. Anders, “Moderated estimation of fold change and dispersion for RNA-seq data,” Genome Biology, vol. 15, 2014.

[23] S. Anders and W. Huber, “Differential expression analysis for sequence count data,” Genome Biology, vol. 11, 2010.

[24] M. D. Robinson, D. J. McCarthy, and G. K. Smyth, “edgeR: A Bioconductor package for differential expression analysis,” Bioinformatics, vol. 26, no. 1, pp. 139–140, 2010.

[25] J. T. Leek, R. B. Scharpf, H. C. Bravo, D. Simcha, B. Langmead, W. E. Johnson, D. Geman, K. Baggerly, and R. A. Irizarry, “Tackling the widespread and critical impact of batch effects in high-throughput data,” Nature Reviews Genetics, vol. 11, no. 10, pp. 733–739, 2010.

[26] A. Subramanian, P. Tamayo, V. K. Mootha, S. Mukherjee, and others, “Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles,” Proceedings of the National Academy of Sciences of the USA, vol. 102, no. 43, pp. 15545–15550, 2005.

[27] S. Monti, P. Tamayo, J. Mesirov, and T. Golub, “Consensus clustering: A resampling-based method for class discovery,” Machine Learning, vol. 52, pp. 91–118, 2003.

[28] V. Vapnik, Statistical Learning Theory, Wiley, 1998.

[29] R. Tibshirani, G. Walther, and T. Hastie, “Estimating the number of clusters in a data set via the gap statistic,” Journal of the Royal Statistical Society, Series B, vol. 63, pp. 411–423, 2001.

[30] M. K. Kerr and G. A. Churchill, “Statistical design and the analysis of gene expression microarray data,” Proceedings of the National Academy of Sciences of the USA, vol. 98, no. 4, pp. 1921–1926, 2001.

You can download the Project files here: Download files now. (You must be logged in).