1.1 Definition of gene expression
Gene expression is the process by which information from a gene is used to synthesize functional gene products, typically proteins or RNA molecules. The level of expression reflects the abundance of transcripts produced from a gene at a given time and under specific conditions, and is a central measure in molecular biology.
1.2 Central dogma of molecular biology
The central dogma describes the flow of genetic information within a biological system: DNA is transcribed into messenger RNA (mRNA), and mRNA is translated into protein. Gene expression data primarily quantifies the mRNA intermediate, capturing the transcriptional activity that precedes protein synthesis. This framework underlies the interpretation of expression measurements.
1.3 Transcriptome and its measurement
The transcriptome is the complete set of RNA transcripts produced by a genome under specific circumstances. It includes mRNA, non-coding RNA, and small RNA species. High-throughput technologies measure the transcriptome by quantifying transcript abundance, providing a snapshot of gene activity. This data is fundamental for studying cellular states, developmental stages, and disease conditions.
2.1 Microarray technology
2.1.1 Principle of hybridization
Microarrays rely on the specific binding (hybridization) of fluorescently labeled cDNA or cRNA to complementary DNA probes immobilized on a solid surface. The intensity of fluorescence at each probe spot is proportional to the abundance of the corresponding transcript in the sample. This method allows parallel quantification of thousands of genes.
2.1.2 Probe design and array types
Probes are short oligonucleotides (or longer cDNA fragments) designed to be unique to target transcripts. Common array platforms include Affymetrix (with multiple perfect-match and mismatch probes per gene), Illumina BeadArrays (with bead-based probes), and custom spotted arrays. Design choices affect specificity, sensitivity, and ability to detect splice variants.
2.2 RNA sequencing (RNA-seq)
2.2.1 Library preparation and sequencing
RNA-seq involves converting RNA into cDNA libraries through reverse transcription, fragmentation, adapter ligation, and PCR amplification. These libraries are sequenced on platforms like Illumina, producing millions of short reads. Paired-end sequencing improves mapping accuracy and detection of splice junctions.
2.2.2 Read alignment and quantification
Raw sequencing reads are aligned to a reference genome or transcriptome using tools such as STAR, HISAT2, or Salmon. Quantification methods (e.g., featureCounts, RSEM) count reads per gene or transcript, generating expression matrices. These counts are later normalized for between-sample comparisons.
2.3 Emerging single-cell technologies
2.3.1 Single-cell RNA-seq (scRNA-seq)
scRNA-seq profiles gene expression in individual cells, revealing cellular heterogeneity within tissues. Techniques such as Drop-seq, 10x Genomics Chromium, and Smart-seq2 isolate and barcode single cells before sequencing. Data from scRNA-seq is sparse and requires specialized preprocessing and analysis methods.
2.3.2 Spatial transcriptomics
Spatial transcriptomics combines gene expression measurement with spatial location information. Methods like MERFISH, seqFISH, and Visium preserve tissue architecture, enabling mapping of transcript abundance onto histological sections. This technology provides insights into tissue organization, cell-cell interactions, and microenvironmental effects.
3.1 Raw data formats
3.1.1 CEL files (microarray)
CEL files store raw probe-level intensity data from Affymetrix arrays. They contain pixel intensities for each probe cell, along with metadata about the scan. Preprocessing extracts and summarizes these intensities into gene-level expression values after background correction and normalization.
3.1.2 FASTQ and SAM/BAM (RNA-seq)
FASTQ files store raw sequencing reads and quality scores. After alignment, reads are stored in SAM (Sequence Alignment/Map) or its binary equivalent BAM format. These aligned files are used for counting reads per genomic feature and for visualizing coverage.
3.2 Normalization methods
3.2.1 Quantile normalization
Quantile normalization makes the distribution of intensities (or counts) identical across samples. It ranks values within each sample, then replaces each rank with the mean of values at that rank across samples. This method is commonly used for microarray data and for RNA-seq data after transformation (e.g., log2 counts per million).
3.2.2 RPKM/FPKM and TPM (RNA-seq)
RPKM (reads per kilobase per million) and FPKM (fragments per kilobase per million) normalize counts by gene length and sequencing depth. TPM (transcripts per million) is a similar measure that scales counts within each sample to sum to one million, making it preferable for cross-sample comparisons. These metrics facilitate comparing expression levels within and between samples, though they assume uniform read distribution along transcripts.
3.2.3 DESeq2 and edgeR scaling
DESeq2 and edgeR use statistical models to estimate size factors or normalization factors based on the assumption that most genes are not differentially expressed. These methods compute a scaling factor that accounts for differences in library size and composition bias, providing robust normalization for differential expression analysis.
3.3 Batch effect correction
3.3.1 Combat and limma
Combat uses an empirical Bayes method to adjust for batch effects by modeling additive and multiplicative batch shifts in expression data. Limma can include batch as a covariate in linear models. Both methods are widely applied to microarray and bulk RNA-seq data to remove technical variation while preserving biological signal.
3.3.2 Harmony and scVI (single-cell)
Harmony uses a soft clustering approach to align single-cell datasets across batches, adjusting principal component (PC) coordinates. scVI is a deep learning–based method that models count data with a variational autoencoder, learning batch-invariant latent representations. These tools address complex batch effects in high-dimensional single-cell expression data.
4.1 Differential expression analysis
4.1.1 Fold change and p-value
Fold change quantifies the magnitude of expression difference between two conditions (e.g., log2 ratio). The p-value assesses the statistical significance of the difference, typically derived from a test such as the t-test, negative binomial test, or Wald test. Both measures are combined to rank candidate genes.
4.1.2 Multiple testing correction
When testing thousands of genes, the probability of false positives increases. Methods like Bonferroni correction, Benjamini-Hochberg false discovery rate (FDR), and q-value adjust p-values to control the expected proportion of false discoveries. Adjusted p-values below a threshold (e.g., 0.05) indicate significant differential expression.
4.1.3 Tools: DESeq2, limma, edgeR
DESeq2 models read counts with a negative binomial distribution and estimates dispersion using shrinkage. limma uses empirical Bayes moderation of variance, applicable to both microarray and RNA-seq data after appropriate transformations. edgeR also employs negative binomial models with robust dispersion estimation. These tools provide comprehensive pipelines for identifying differentially expressed genes.
4.2 Clustering and dimensionality reduction
4.2.1 Principal component analysis (PCA)
PCA reduces the dimensionality of expression data by transforming correlated variables into a set of orthogonal principal components that capture the largest variance. It is commonly used for data exploration, detecting outliers, and visualizing sample relationships in a low-dimensional space.
4.2.2 t-SNE and UMAP
t-SNE (t-distributed stochastic neighbor embedding) and UMAP (uniform manifold approximation and projection) are non-linear dimensionality reduction methods that preserve local structure. They are particularly useful for visualizing single-cell or large-scale transcriptomic data, revealing clusters of cells or samples with similar expression profiles.
4.2.3 Hierarchical clustering and k-means
Hierarchical clustering builds a tree of relationships based on pairwise distances (e.g., Euclidean, correlation) and can be visualized as a dendrogram. k-means partitions samples or genes into a predefined number of clusters by minimizing within-cluster variance. Both methods are used to identify groups of co-expressed genes or sample subtypes.
4.3 Time-series analysis
4.3.1 Temporal expression profiles
Time-series gene expression data are collected at multiple time points to study dynamic processes such as development, circadian rhythms, or drug response. Statistical methods (e.g., spline fitting, autoregressive models) are used to model expression trajectories over time.
4.3.2 Gene expression dynamics
Analysis of time-series data can identify genes with periodic, monotonic, or transient patterns. Clustering of temporal profiles reveals groups of genes that share similar dynamics, which may indicate co-regulation or functional coordination during the observed process.
5.1 Gene ontology enrichment
5.1.1 Over-representation analysis (ORA)
ORA tests whether a set of differentially expressed genes contains more genes associated with a particular Gene Ontology (GO) term than expected by chance. A hypergeometric or Fisher’s exact test is applied, and multiple testing correction is used to identify significantly enriched biological processes, molecular functions, or cellular components.
5.1.2 Gene set enrichment analysis (GSEA)
GSEA evaluates whether a predefined gene set is enriched at the top or bottom of a ranked list of all genes (e.g., ranked by fold change). It uses a running sum statistic and permutation testing, and does not require arbitrary cutoffs for differential expression. This method can detect coordinate changes in gene sets even if individual genes are not significantly altered.
5.2 Pathway analysis
5.2.1 KEGG and Reactome
KEGG (Kyoto Encyclopedia of Genes and Genomes) maps genes to metabolic and signaling pathways, providing a rich set of curated pathway diagrams. Reactome is an open-source, curated database of human biological reactions and pathways. Both databases support enrichment and visualization tools, helping researchers interpret expression changes in a pathway context.
5.2.2 Protein-protein interaction networks
Protein-protein interaction (PPI) networks connect proteins known to physically interact, often sourced from databases like STRING or BioGRID. Integrating differential expression data with PPI networks can identify interaction hubs and subnetworks that are dysregulated in a condition, offering mechanistic insights beyond single-gene analysis.
5.3 Regulatory network inference
5.3.1 Co-expression networks (WGCNA)
Weighted Gene Co-expression Network Analysis (WGCNA) constructs networks based on pairwise correlations of gene expression across samples. Genes with similar expression patterns are grouped into modules, which can be associated with sample traits (e.g., disease status). Module hubs and eigengenes highlight regulatory relationships.
5.3.2 Transcription factor binding motifs
Transcription factor (TF) binding motifs are short DNA sequences recognized by TFs. By scanning promoters or enhancer regions of co-expressed gene sets for overrepresentation of known motifs, researchers can infer which TFs likely regulate those genes. Tools like HOMER or MEME are used for this analysis, linking expression patterns to upstream regulators.
6.1 Public data archives
6.1.1 Gene Expression Omnibus (GEO)
GEO, maintained by NCBI, is the largest public repository for high-throughput functional genomics data. It accepts microarray, RNA-seq, and other expression data, along with metadata. Users can query, download, and reanalyze datasets, making it a cornerstone of open science in genomics.
6.1.2 ArrayExpress
ArrayExpress, hosted by EMBL-EBI, is a similar database for gene expression and other functional genomics data. It supports MIAME-compliant submissions and provides tools for data querying and analysis. Cross-linking with GEO facilitates data integration.
6.1.3 Sequence Read Archive (SRA)
SRA stores raw sequencing data from high-throughput sequencing projects, including RNA-seq. It is often used as the primary source for read-level data, while processed expression matrices may be deposited in GEO or ArrayExpress. SRA provides unique accessions for reproducibility and reanalysis.
6.2 Specialized databases
6.2.1 GTEx (tissue expression)
The Genotype-Tissue Expression (GTEx) project collects RNA-seq data from multiple tissues of postmortem donors. It provides a comprehensive resource for studying tissue-specific expression patterns, expression quantitative trait loci (eQTLs), and gene regulation across human tissues.
6.2.2 TCGA (cancer genomics)
The Cancer Genome Atlas (TCGA) provides genomic, transcriptomic, and clinical data for over 30 cancer types. Its RNA-seq and microarray data have been used extensively for cancer subtype classification, biomarker discovery, and survival analysis. TCGA data are accessible through portals like GDC.
6.2.3 Human Cell Atlas (single-cell)
The Human Cell Atlas is a large-scale initiative to create a reference map of all human cell types based on single-cell RNA-seq and spatial data. It aims to characterize cellular diversity across tissues and developmental stages, serving as a foundational resource for understanding human biology and disease.
7.1 Disease biomarker discovery
Gene expression data are used to identify genes or gene signatures that are differentially expressed between healthy and diseased individuals. Such biomarkers can aid early diagnosis, prognosis, or monitoring of disease progression. For example, expression panels are used in oncology to guide treatment decisions.
7.2 Cancer subtype classification
Transcriptomic profiling has revealed molecular subtypes within cancers (e.g., breast cancer Luminal A/B, HER2-enriched, basal-like) that have distinct clinical outcomes and responses to therapy. Clustering and classification algorithms applied to expression data enable precision oncology and personalized treatment strategies.
7.3 Pharmacogenomics and drug response
Expression data help predict how individuals or tumors respond to drugs. Genes involved in drug metabolism, efflux, or target pathways can be correlated with sensitivity or resistance. This information informs drug development, repurposing, and the design of combination therapies.
7.4 Developmental and evolutionary biology
Comparing gene expression across developmental stages sheds light on the molecular programs driving embryogenesis, cell differentiation, and organogenesis. Cross-species expression comparisons (e.g., using phylogenetically informed methods) reveal conserved and divergent regulatory mechanisms, contributing to understanding evolution of gene expression.