RNA Sequencing Analysis: From FASTQ Files to Biological Questions

If you are a bench biologist, a bioinformatics beginner, or a seasoned researcher moving into transcriptomics, this guide gives you a structured and flexible approach to RNA sequencing analysis. It does not prescribe one universal pipeline because no single workflow fits every study. Instead, it helps you design an analysis plan, select a suitable workflow, inspect data quality at each step, and document your decisions with transparency.

At a Glance

Decision Criteria for Your RNA seq Workflow

The first fork in the road is whether you need to discover novel transcripts or quantify known ones. For most differential expression studies using a well annotated reference genome, quantification by alignment (STAR followed by featureCounts) or pseudoalignment (Salmon or Kallisto) works well. Pseudoalignment is faster and less memory intensive, but some downstream tools require aligned BAM files for visualization or splice event detection.

Choose stranded library preparation if possible. Strandedness dramatically improves annotation accuracy and reduces false positives in differential expression. If you use unstranded data, document this choice and apply appropriate counting parameters.

The number of biological replicates strongly influences your ability to detect true differences. Three or more replicates per condition are common, but you should check the power requirements for your expected effect sizes and expression variability using tools like RNASeqPower or Scotty.

Practical Workflow and Implementation Sequence

Step 1: Quality Control and Trimming

Start by running FastQC on your raw FASTQ files. Look for per base quality scores below 30, adapter contamination, and GC bias. Aggregate all sample reports with MultiQC to get a global view.

If adapters are present, trim them using a tool like Fastp or Trimmomatic. The Galaxy Training Network offers excellent tutorials for running these tools in a graphical interface. Galaxy also provides preconfigured workflows that you can adapt.

Step 2: Read Alignment or Quantification

For alignment based methods, use a splice aware aligner such as STAR or HISAT2. Build your index from the same genome and annotation version you plan to use for counting.

For pseudoalignment, Salmon or Kallisto directly quantify transcript abundance without producing a BAM file. They are fast but sacrifice some positional information. If you later need to visualize aligned reads on a genome browser, you must realign with STAR.

The NCBI Sequence Read Archive is an excellent resource to download published RNA seq data for testing or replication. When you publish your own data, deposit it there with detailed metadata.

Step 3: Expression Quantification

For alignment based data, use featureCounts to obtain gene level counts. For pseudoalignment, use the built in quant functions. Normalize counts using TPM (transcripts per million) for within sample comparisons, and use raw counts for differential expression tools that model count distributions.

Check the proportion of reads that mapped to exons versus introns or intergenic regions. High intronic reads may indicate incomplete splicing or DNA contamination. Document any filtering thresholds.

Step 4: Differential Expression Analysis

Use DESeq2, edgeR, or limma voom from Bioconductor. These packages are well documented and widely tested. They work with raw count matrices and apply statistical models that account for library size and dispersion.

Set a significance threshold and apply multiple testing correction (Benjamini Hochberg is standard). Report both p values and adjusted p values, and provide fold change estimates with confidence intervals if possible.

Step 5: Functional Interpretation

Enrichment analysis with tools like clusterProfiler or GSEA helps you move from a list of differentially expressed genes to biological processes. Use gene set collections that are relevant to your organism and tissue type. The EMBL EBI Training portal offers courses on functional annotation and interpretation.

Do not stop at enrichment p values. Look at the direction of change, the magnitude of fold changes, and the overlap between pathways. Always validate key findings with an independent method such as qPCR or western blot.

Step 6: Documentation and Reproducibility

Record every software version, parameter setting, and reference file used. Keep a lab notebook or use a computational workflow management tool like Snakemake or Nextflow. The NCBI Bookshelf contains chapters on best practices for reproducible bioinformatics research.

Common Mistakes

One frequent error is ignoring strandness when setting up alignment or counting. If you specify the wrong library type, you may lose half of your reads.

Another mistake is over filtering. Removing too many low count genes reduces your power to detect differential expression. Use a per condition filter (e.g., at least 10 counts in at least 3 samples) rather than a global threshold.

Batch effects are often overlooked until late in analysis. Record the date of library preparation, the technician, and the sequencing lane. Use tools like RUVseq or ComBat to check for and correct batch effects. Our related guide on batch effects provides a deeper treatment.

Finally, do not skip the exploratory data analysis step. PCA plots and heatmaps of the top variable genes can reveal sample outliers, batch structure, or unexpected clustering.

Limits and Uncertainty

RNA seq analysis is powerful but not perfect. You cannot equate expression levels measured in TPM or FPKM across experiments because normalization differs. Reads may multi map to paralogous genes, causing ambiguous counts. Lowly expressed genes are noisy and often yield false positives in differential testing.

For non model organisms, the reference genome and annotation may be incomplete. In that case, consider de novo transcriptome assembly with tools like Trinity. Be aware that isoform quantification is less reliable than gene level quantification.

Single cell RNA seq adds another layer of complexity with drop out zeros and high technical noise. Our post on single cell RNA seq workflow covers that topic in detail.

Always interpret differential expression results in the context of your biological question. A statistically significant gene may have a small fold change that is not biologically meaningful. Conversely, a gene with a large fold change but borderline p value may still be worth investigating if it belongs to a relevant pathway.

Frequently Asked Questions

1. Should I use genome alignment or pseudoalignment?

It depends on your goals. For standard differential expression with a well annotated reference, pseudoalignment is faster and uses less memory. Choose alignment if you need to detect splice junctions, fusion transcripts, or perform variant calling from RNA seq data.

2. How many replicates do I need for RNA seq?

Three biological replicates per condition is a common minimum. For high variability tissues or subtle expected changes, five to ten may be required. Use power analysis tools to estimate the optimal number for your specific experimental design.

3. What is the best normalization method for RNA seq data?

For differential expression, use tools that model count distributions (DESeq2, edgeR) as they account for library size and composition effects themselves. TPM is useful for within sample comparisons but should not be used across samples for DE analysis without proper statistical handling.

4. How do I handle samples with poor quality metrics?

If a sample has very low mapped read percentage or high duplication, you can either drop it or try to rescue with additional filtering. Document your decision. Dropping a sample reduces power but may be necessary. If you keep it, check whether it clusters with its biological group in PCA.

References and Further Reading

• NCBI Bookshelf provides authoritative chapters on RNA seq analysis and bioinformatics fundamentals.
• EMBL EBI Training offers free courses on sequence analysis, functional interpretation, and data management.
• Galaxy Training Network features hands on tutorials for RNA seq workflows in a user friendly web interface.
• Bioconductor hosts the most widely used R packages for RNA seq analysis with full documentation.
• NCBI Sequence Read Archive is the primary repository for raw sequencing data. Use it to download public datasets for testing.
• A study on mevalonate pathway dysregulation in pancreatic cancer used bulk RNA seq to define transcriptomic progression markers (PubMed 42436522).
• Single cell and spatial transcriptomic analysis of glioblastoma revealed lactate driven crosstalk using RNA seq workflows (PubMed 42436508).
• Cross species single cell RNA seq helped elucidate T cell roles in aortic dissection pathogenesis (PubMed 42436415).

Related Articles

• Single Cell RNA seq Workflow: A Practical Analysis Roadmap
• Single Cell RNA seq Quality Control: Cells, Genes, and Mitochondrial Reads
• Cell Type Annotation in Single Cell RNA seq: A Layered Evidence Framework
• Batch Effects in RNA seq: How to Recognize and Plan Around Them
• Pseudobulk Analysis for Single Cell RNA seq