RNA-Seq Normalization Methods: TPM, RPKM, and Beyond

RNA sequencing produces raw read counts that reflect both biological expression levels and technical artifacts such as sequencing depth, gene length, and composition biases. Normalization methods adjust these raw counts so that expression measures can be compared within a sample and across samples. This article explains the principles behind TPM, RPKM/FPKM, TMM, and DESeq2 median-of-ratios normalization, and provides practical guidance for selecting methods based on analytical goals. The intended readers are students, researchers, analysts, and life-science professionals who need to make informed decisions about RNA-seq data processing.

Why Normalization Matters in RNA-Seq Analysis

Raw RNA-seq data consist of counts that reflect how many sequencing reads mapped to each gene. These counts are influenced by multiple factors unrelated to true biological expression. Library size varies between samples because sequencing machines produce different total numbers of reads per lane. Gene length affects counts because longer genes generate more fragments than shorter genes at equal expression levels. Composition effects arise when a small number of highly expressed genes consume a large fraction of the sequencing capacity, suppressing counts for other genes.

Normalization is the step that adjusts raw data to account for factors that prevent direct comparison of expression measures. Errors in normalization can have a significant impact on downstream analysis, such as inflated false positives in differential expression analysis. The choice of normalization method should be guided by the assumptions each method makes about the data and whether those assumptions hold for the specific experiment. Selecting a normalization method with assumptions that are met and that produces a meaningful measure of expression is necessary for valid biological conclusions.

The practical consequence of poor normalization is that differentially expressed genes may be identified incorrectly. A gene may appear upregulated simply because its sample had more total reads. Another gene may appear downregulated because a few highly expressed genes dominated the sequencing capacity. Normalization methods attempt to correct these distortions, but each method corrects them differently and makes different assumptions about what should remain constant across samples.

At a Glance

Core Principles of Read Count Normalization

Sequencing Depth and Library Size

The most basic adjustment in RNA-seq normalization accounts for the total number of reads sequenced per sample. A sample sequenced to 50 million reads will have roughly twice the counts of a sample sequenced to 25 million reads for the same biological material. Counts per million (CPM) divides each gene count by the total library size and multiplies by one million. This simple approach makes samples comparable in terms of sequencing effort.

Library size normalization assumes that the total RNA amount is similar across samples and that most genes are not differentially expressed. When these assumptions hold, dividing by total counts is a reasonable adjustment. However, when a small number of genes are highly expressed in one condition but not another, the total count approach can introduce bias. The highly expressed genes consume more of the sequencing capacity in one sample, making all other genes appear reduced in that sample.

Gene Length and Transcript Length

Gene length normalization addresses the fact that longer transcripts produce more reads than shorter transcripts at the same molar concentration. A 4,000 base pair gene will generate approximately twice the reads of a 2,000 base pair gene if both are expressed at the same level. This adjustment is essential when comparing expression levels between genes within a sample. It is less critical when comparing the same gene across samples, because gene length is constant for a given gene.

The reads per kilobase per million mapped reads (RPKM) metric divides counts by gene length in kilobases and by the total number of mapped reads in millions. The fragments per kilobase per million mapped reads (FPKM) metric is the paired-end equivalent that counts fragments instead of reads. These metrics allow within-sample comparison of expression levels across genes of different lengths.

Composition Bias and Effective Library Size

Composition bias occurs when a subset of genes changes expression dramatically between conditions. For example, in a study comparing liver and muscle tissue, genes specific to each tissue will be highly expressed in their respective samples. These tissue-specific genes consume sequencing capacity, making other genes appear less abundant. Methods that use total library size fail to correct this bias because the total count is dominated by the differentially expressed genes.

Effective library size methods estimate the library size that would be observed if composition bias were removed. The trimmed mean of M values (TMM) method from edgeR and the median-of-ratios method from DESeq2 both estimate scaling factors that adjust library sizes to account for composition effects. These methods assume that most genes are not differentially expressed and use the stable majority of genes to estimate the scaling factor.

Within-Sample Normalization Metrics

RPKM and FPKM

RPKM normalizes for both sequencing depth and gene length. The calculation divides the read count by the gene length in kilobases and by the total number of mapped reads in millions. FPKM is the fragment-based version used for paired-end sequencing data. These metrics are appropriate for comparing expression levels between genes within a single sample, such as identifying which genes are most highly expressed in a tissue.

The limitation of RPKM and FPKM is that they do not correct for composition bias. The total number of mapped reads includes reads from highly expressed genes that may differ between samples. When comparing the same gene across samples with different compositions, RPKM and FPKM values can be misleading. The sum of RPKM values across all genes in a sample is not constant, which complicates comparisons between samples.

TPM

Transcripts per million (TPM) addresses the composition bias problem in within-sample normalization. The calculation first divides read counts by gene length to obtain rates per transcript. These rates are then divided by the sum of all rates in the sample and multiplied by one million. The result is that TPM values sum to one million across all genes in each sample, making the metric directly comparable across samples.

The key difference between TPM and RPKM is the order of operations. RPKM divides by total reads first and then by gene length. TPM divides by gene length first and then normalizes by the total rate. This ordering means that TPM values are not affected by the average transcript length of the sample, while RPKM values are. For cross-sample comparisons, TPM is generally preferred over RPKM because the sum of TPM values is constant across samples.

When to Use Within-Sample Metrics

TPM is appropriate for comparing expression levels across samples when the goal is to identify relative expression changes. It is also useful for visualizing expression patterns in heatmaps and for methods that assume comparable distributions across samples. RPKM and FPKM remain in use but are less suitable for cross-sample comparisons due to composition sensitivity.

For differential expression analysis, TPM values are often used as input to statistical methods that model count distributions. However, the statistical methods themselves typically perform their own normalization steps. The choice of TPM versus raw counts depends on the analysis pipeline and the assumptions of the downstream statistical tests.

Between-Sample Normalization Methods

TMM from edgeR

The trimmed mean of M values method estimates a scaling factor for each sample relative to a reference sample. The method computes log fold changes (M values) and absolute expression levels (A values) for each gene between a sample and the reference. It then trims the most extreme M values and A values and calculates a weighted average of the remaining values. This trimmed average is the scaling factor for the sample.

TMM assumes that most genes are not differentially expressed between samples. The trimming removes genes with extreme fold changes and extreme expression levels, leaving the stable majority of genes to estimate the scaling factor. This approach is robust to composition bias because the highly expressed and highly variable genes are excluded from the calculation.

The TMM method is implemented in the edgeR Bioconductor package and is widely used for differential expression analysis. The scaling factors are applied to the library sizes before fitting the statistical model. TMM is appropriate for bulk RNA-seq data with moderate to large numbers of samples and for experiments where most genes are expected to be unchanged.

DESeq2 Median-of-Ratios

The median-of-ratios method estimates a size factor for each sample by computing the median of the ratios of each gene count to the geometric mean of that gene across all samples. The geometric mean serves as a pseudo-reference sample. Genes with zero counts in any sample are excluded from the calculation. The median of these ratios is the size factor for the sample.

This method assumes that most genes are not differentially expressed and that the counts follow a negative binomial distribution. The size factors account for both sequencing depth and composition bias. DESeq2 applies these size factors within its generalized linear model framework for differential expression testing.

The median-of-ratios method is robust when the assumption of mostly unchanged genes holds. It performs reasonably well even when some genes are differentially expressed, as long as the majority remain stable. The method is implemented in the DESeq2 Bioconductor package and is one of the most commonly used normalization approaches for bulk RNA-seq.

Upper Quartile Normalization

Upper quartile normalization scales each sample by the 75th percentile of its count distribution. This method assumes that the upper quartile of genes is relatively stable across samples and that changes in expression are concentrated in a subset of genes. The scaling factor is the ratio of the sample upper quartile to a reference upper quartile.

A comparison of normalization methods for TempO-Seq targeted RNA sequencing data found that upper quartile normalization performed best for maintaining fold change levels in two-group comparisons. The specificity of upper quartile normalization was greater than 0.84 and sensitivity greater than 0.90 for most fold change levels tested. The method also agreed most closely with simulated fold change assignments in clustering analysis.

Upper quartile normalization is particularly useful for targeted sequencing panels where a sentinel gene set represents genome-wide transcription. It is less commonly used for whole transcriptome analysis but can be appropriate when the assumption of stable upper quartile expression holds.

Relative Log Expression from DESeq2

The relative log expression (RLE) method is closely related to the median-of-ratios approach. It computes the ratio of each gene count to the geometric mean across samples, then takes the median of these ratios for each sample. The RLE method is the default normalization in DESeq2 and produces size factors equivalent to the median-of-ratios approach.

A comparison of normalization methods using Shannon entropy across multiple cancer types found that TPM, RLE, and TMM normalization coupled with a log2 fold change threshold of at least 1 yielded the best results for identifying differentially expressed genes. This finding supports the use of these methods for standard differential expression workflows.

Selecting the Appropriate Normalization Method

Analytical Goals and Method Selection

The choice of normalization method depends on the analytical goal. For differential expression analysis, methods that estimate effective library sizes such as TMM and DESeq2 median-of-ratios are generally preferred because they correct for composition bias. For comparing expression levels across genes within a sample, TPM is appropriate because it accounts for gene length and produces a constant sum across samples.

For cross-study integration and machine learning applications, the choice of preprocessing operations affects classifier performance. A comparison of RNA-seq data preprocessing pipelines for transcriptomic predictions across independent studies found that batch effect correction improved performance when training on TCGA data and testing on GTEx data. The use of preprocessing operations sometimes worsened classification performance when the independent test dataset was aggregated from separate sources.

For single-cell RNA-seq data, normalization considerations differ from bulk RNA-seq due to low input material, dropout events, and various forms of bias present in the sequencing process. Methods designed specifically for single-cell data may express advantages over bulk methods depending on available information and data type. The choice of normalization method affects downstream analysis such as classification or differential analysis.

Sample Size and Experimental Design

The number of samples per group influences the performance of normalization methods. An evaluation of RNA-seq differential analysis methods found that with three samples per group, EBSeq performed better than other methods in terms of false discovery rate control, power, and stability for negative binomial distributed data. When sample sizes increased to six or twelve per group, DESeq2 performed slightly better than other methods. All methods improved performance when sample size increased to twelve per group except DESeq.

For experiments without replicates, the choice of normalization method is constrained. A comparison of TMM, RLE, and MRN normalization methods for a two-condition without replicates design showed mathematical conditions under which these methods give exactly the same results. Understanding these equivalences helps analysts interpret results from different tools.

Data Characteristics and Assumption Checking

The assumptions underlying each normalization method should be checked before application. TMM and DESeq2 assume that most genes are not differentially expressed. If this assumption is violated, such as in experiments comparing very different cell types or tissues, the scaling factors may be biased. Upper quartile normalization assumes stability of the upper quartile of expression. TPM assumes that total transcript content is similar across samples.

An evaluation of normalization methods for differential expression analysis found that per-gene normalization approaches after per-sample global scaling achieved high specificity and detection power on benchmark datasets. The commonly used methods DESeq and TMM yielded higher power but traded off with reduced specificity. These tradeoffs should be considered when selecting methods for a specific experiment.

Practical Workflow for Normalization

Step 1: Assess Data Quality

Before applying normalization, examine the raw count data for quality issues. Check the total number of reads per sample and identify any samples with unusually low library sizes. Examine the distribution of counts across genes and identify any genes with zero counts in all samples. Assess whether the number of samples per group meets the minimum requirements for the chosen statistical methods.

Quality assessment should include examination of alignment statistics, duplication rates, and the proportion of reads mapping to exonic regions. Samples with poor quality metrics should be flagged and potentially excluded from the analysis. The choice of normalization method cannot compensate for fundamentally poor data quality.

Step 2: Choose the Normalization Method Based on Analysis Goal

For differential expression analysis with bulk RNA-seq data and multiple samples per group, use TMM from edgeR or median-of-ratios from DESeq2. These methods correct for composition bias and are robust when most genes are unchanged. For experiments with very different sample compositions, consider upper quartile normalization as an alternative.

For comparing expression levels across genes within a sample, use TPM. This metric accounts for gene length and produces comparable values across samples. For cross-study integration, consider batch effect correction in addition to normalization. The choice of preprocessing operations should be validated on the specific data and prediction task.

Step 3: Apply Normalization and Verify Results

Apply the chosen normalization method using the appropriate software package. After normalization, verify that the scaling factors are reasonable. Extreme scaling factors may indicate problems with library preparation or data quality. Examine the distribution of normalized values across samples to confirm that they are comparable.

Check whether the normalization successfully removed technical variation while preserving biological differences. Principal component analysis or clustering of normalized data should show separation by biological condition instead of by sequencing batch or library size. If samples cluster by technical factors, additional correction may be needed.

Step 4: Document Normalization Parameters

Record the normalization method, software version, and parameters used in the analysis. This documentation is essential for reproducibility and for interpreting results in the context of the chosen methods. Include the scaling factors or size factors in the analysis records so that other researchers can verify the normalization.

The FAIR Guiding Principles emphasize that data should be findable, accessible, interoperable, and reusable. Documenting normalization choices supports these principles by making the analysis transparent and reproducible. The National Institutes of Health Genomic Data Sharing Policy requires that data sharing plans address the handling of genomic data, including analysis methods.

Records and Measurements for Normalization

Key Metrics to Record

Maintain records of the following metrics for each RNA-seq dataset:

• Total number of reads sequenced per sample
• Number of reads mapped to the reference genome
• Percentage of reads mapped to exonic, intronic, and intergenic regions
• Number of genes with nonzero counts in each sample
• Library size before and after normalization
• Scaling factors or size factors estimated by the normalization method
• Software versions and parameters used for normalization

These records allow other researchers to reproduce the analysis and to assess whether the normalization choices were appropriate. The records also support troubleshooting if downstream analysis produces unexpected results.

Quality Control Thresholds

Quality control thresholds should be established before analysis and applied consistently across samples. Common thresholds include minimum total reads per sample, minimum percentage of mapped reads, and maximum percentage of reads mapping to ribosomal RNA. Samples failing these thresholds should be flagged for review.

The choice of thresholds depends on the sequencing platform, library preparation method, and organism. Thresholds should be documented in the analysis plan and applied consistently. Samples that fail quality control should be excluded from the analysis or re-sequenced if possible.

Verification of Normalization Effectiveness

After normalization, verify that the method achieved its intended effect. Compare the distribution of normalized values across samples. The distributions should be similar if the normalization successfully removed technical variation. Examine whether the scaling factors correlate with library size or other technical variables.

For differential expression analysis, verify that the normalization did not introduce artifacts. Check that the number of differentially expressed genes is reasonable for the experimental design. Examine whether known positive or negative control genes show the expected patterns. If results are unexpected, revisit the normalization choices.

Common Failure Patterns in Normalization

Failure to Account for Composition Bias

A common failure is using total count normalization when composition bias is present. This occurs when a small number of genes are highly expressed in one condition but not another. The total count approach makes all other genes appear reduced in the sample with the highly expressed genes. This pattern can produce false positives in differential expression analysis.

The solution is to use composition-aware methods such as TMM or DESeq2 median-of-ratios. These methods estimate effective library sizes that exclude the influence of highly variable genes. The choice of method should be based on the expected degree of composition bias in the experiment.

Applying Gene Length Normalization for Cross-Sample Comparisons

Another failure pattern is using RPKM or FPKM for cross-sample comparisons without recognizing the composition sensitivity of these metrics. RPKM values do not sum to a constant across samples, so comparisons of the same gene across samples can be misleading. The total number of reads in the denominator includes reads from genes that may differ between samples.

TPM is preferred for cross-sample comparisons because the sum of TPM values is constant across samples. When RPKM or FPKM values are used for cross-sample comparisons, the results should be interpreted with caution and verified with an alternative method.

Ignoring Assumptions of the Normalization Method

Each normalization method makes assumptions about the data. TMM and DESeq2 assume that most genes are not differentially expressed. Upper quartile normalization assumes stability of the upper quartile. TPM assumes similar total transcript content across samples. When these assumptions are violated, the normalization method may perform poorly.

Researchers should assess whether the assumptions hold for their specific experiment. For experiments comparing very different cell types or conditions, the assumption of mostly unchanged genes may be violated. In these cases, alternative methods or additional corrections may be needed.

Using Inappropriate Methods for Single-Cell Data

Single-cell RNA-seq data have characteristics that differ from bulk RNA-seq, including low input material, dropout events, and high technical variability. Normalization methods designed for bulk data may not perform optimally for single-cell data. Methods designed specifically for single-cell sequencing may express advantages depending on available information and data type.

The choice of normalization method for single-cell data should consider the specific platform, the presence of spike-in genes, and the downstream analysis goals. Methods that rely on spike-in genes require that spike-ins were added during library preparation. Methods that do not depend on spike-ins can be used when spike-ins are not available.

Limitations and Interpretation Constraints

Normalization Cannot Correct for Poor Data Quality

Normalization adjusts for technical variation in read counts, but it cannot compensate for fundamentally poor data quality. Samples with very low library sizes, high duplication rates, or poor mapping rates will produce unreliable results regardless of the normalization method. Quality control should be performed before normalization, and poor-quality samples should be excluded.

The choice of normalization method should not be used to rescue datasets with severe quality issues. If a sample has an extremely low library size, the scaling factor will be large and may introduce noise into the analysis. Excluding such samples is preferable to including them with extreme scaling factors.

Normalization Methods Make Different Assumptions

The assumptions underlying normalization methods are not always stated explicitly, but they have substantial impact on performance. Methods perform poorly when their assumptions are violated, and this causes problems in subsequent analysis. Understanding the assumptions is necessary for choosing methods appropriate for the data at hand.

For example, TMM and DESeq2 assume that most genes are unchanged between conditions. If this assumption is violated, the scaling factors will be biased and differential expression results will be unreliable. Researchers should assess whether the assumption of mostly unchanged genes is reasonable for their experiment.

Interpretation of Normalized Values

Normalized values such as TPM and RPKM are relative measures of expression, not absolute quantities. A TPM value of 100 means that the gene accounts for 100 transcripts per million transcripts in the sample. This value is useful for comparing relative expression but does not indicate the absolute number of transcripts per cell.

The interpretation of normalized values depends on the normalization method and the assumptions it makes. Values from different normalization methods are not directly comparable. When reporting results, the normalization method should be stated clearly so that readers can interpret the values appropriately.

Cross-Study Comparisons Require Careful Consideration

Comparing normalized values across studies is challenging because different studies may use different normalization methods, reference genomes, and analysis pipelines. A comparison of RNA-seq data preprocessing pipelines found that the choice of preprocessing operations affected the performance of classifier models for tissue of origin predictions. Batch effect correction improved performance in some cases but worsened it in others.

For cross-study comparisons, researchers should use consistent preprocessing methods and consider batch effect correction. The FAIR Guiding Principles support the reuse of data by making data findable, accessible, interoperable, and reusable. However, the choice of normalization method should be documented and considered when interpreting cross-study results.

Safety and Regulatory Context for Genomic Data

Data Sharing Requirements

Genomic data generated from human subjects are subject to data sharing policies that address privacy and confidentiality. The National Institutes of Health Genomic Data Sharing Policy requires that researchers plan for the sharing of genomic data generated with NIH funding. The policy addresses informed consent, data security, and the conditions under which data can be shared.

Researchers should be aware of these requirements when planning RNA-seq experiments and when sharing normalized data. The normalization method and analysis parameters should be documented to support data sharing and reuse. Data repositories such as the National Center for Biotechnology Information provide infrastructure for sharing genomic data.

Reproducibility and Documentation

Reproducibility requires that the analysis pipeline, including normalization choices, be documented in sufficient detail for other researchers to replicate the results. The FAIR Guiding Principles emphasize that data should be findable, accessible, interoperable, and reusable. Documentation of normalization methods supports these principles.

The European Bioinformatics Institute provides training resources for bioinformatics analysis, including RNA-seq data processing. These resources can help researchers understand the principles behind normalization methods and apply them appropriately.

Professional Escalation Criteria

Researchers should escalate to a bioinformatics specialist or statistician when:

• The normalization method produces extreme scaling factors that cannot be explained by library size differences
• The data violate the assumptions of the chosen normalization method in ways that cannot be addressed by alternative methods
• The results of differential expression analysis are inconsistent with known biology or expected patterns
• The analysis involves complex experimental designs that require specialized statistical expertise
• The data will be shared publicly and require compliance with data sharing policies

Early consultation with a specialist can prevent downstream analysis problems and ensure that normalization choices are appropriate for the experimental design.

Frequently Asked Questions

What is the difference between RPKM and TPM?

RPKM divides read counts by gene length in kilobases and by total mapped reads in millions. TPM divides read counts by gene length first, then normalizes by the sum of all rates in the sample and multiplies by one million. The key difference is that TPM values sum to one million across all genes in each sample, making them directly comparable across samples. RPKM values do not have this property because the total reads in the denominator include reads from genes that may differ between samples.

When should I use TMM instead of TPM?

Use TMM for differential expression analysis when you need to correct for composition bias between samples. TMM estimates scaling factors based on the stable majority of genes and is implemented in the edgeR package. Use TPM when you need to compare expression levels across genes within a sample or when you need a normalized value that is comparable across samples for visualization or clustering.

Does DESeq2 perform its own normalization?

DESeq2 estimates size factors using the median-of-ratios method and applies these size factors within its generalized linear model framework. The size factors account for sequencing depth and composition bias. When using DESeq2, you do not need to normalize the data separately before running the differential expression analysis. The size factors are estimated internally and used in the statistical model.

Can I use TPM values as input to DESeq2?

DESeq2 expects raw count data as input because it models the count distribution and estimates size factors from the counts. Using TPM values as input would violate the assumptions of the negative binomial model. If you have TPM values and need to perform differential expression analysis, you should obtain the raw counts and use them as input to DESeq2 or edgeR.

What normalization method is best for single-cell RNA-seq data?

The best normalization method for single-cell RNA-seq data depends on the platform, the presence of spike-in genes, and the downstream analysis goals. Methods designed specifically for single-cell data may express advantages over bulk methods. Some methods rely on spike-in genes, while others do not. The choice should be based on the available information and the type of data.

How do I know if my normalization method is working correctly?

After normalization, examine the distribution of normalized values across samples. The distributions should be similar if the normalization successfully removed technical variation. Check that the scaling factors are reasonable and do not correlate with technical variables other than library size. Perform principal component analysis or clustering to verify that samples group by biological condition instead of by technical factors.

What should I do if my samples have very different library sizes?

If samples have very different library sizes, use a normalization method that accounts for sequencing depth, such as TMM, DESeq2 median-of-ratios, or TPM. These methods estimate scaling factors that adjust for library size differences. However, extremely different library sizes may indicate quality problems, and samples with very low library sizes should be examined for quality issues before inclusion in the analysis.

Can I compare normalized values across different studies?

Comparing normalized values across studies is challenging because different studies may use different normalization methods, reference genomes, and analysis pipelines. The choice of preprocessing operations affects the performance of downstream analyses. For cross-study comparisons, use consistent preprocessing methods and consider batch effect correction. Document the normalization method used so that results can be interpreted appropriately.

Related Bioinformatics Guides

• RNA-Seq Differential Expression: DESeq2, edgeR, and limma-voom Frameworks
• Single-Cell RNA-Seq Normalization: Batch Effect Correction and Dimension Reduction (PCA, t-SNE, UMAP)
• Alternative Splicing Analysis from RNA-Seq Data
• Single-Cell RNA Sequencing: From Bulk to Resolution
• Master Guide: Single-Cell RNA Sequencing Bioinformatics Workflows

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Current best practices in single-cell RNA-seq analysis: a tutorial.. Molecular systems biology, 2019.
• Comparison of Normalization Methods for Analysis of TempO-Seq Targeted RNA Sequencing Data.. Frontiers in genetics, 2020.
• Normalization Methods on Single-Cell RNA-seq Data: An Empirical Survey.. Frontiers in genetics, 2020.
• Transcriptional and Cellular Diversity of the Human Heart.. Circulation, 2020.
• Synergistic efficacy of simultaneous anti-TGF-β/VEGF bispecific antibody and PD-1 blockade in cancer therapy.. Journal of hematology & oncology, 2023.
• Selecting between-sample RNA-Seq normalization methods from the perspective of their assumptions.. Briefings in bioinformatics, 2018.
• An evaluation of RNA-seq differential analysis methods.. PloS one, 2022.
• Super-enhancer-driven TOX2 mediates oncogenesis in Natural Killer/T Cell Lymphoma.. Molecular cancer, 2023.
• Multi-meta-omics reveal unique symbiotic synchronization between ectomycorrhizal fungus and soil microbiome in Tricholoma matsutake habitat.. 2025.
• TORC1-dependent translation drives chromatin remodeling during the germ-cell-to-maternal transition in Drosophila.. 2026.
• Assessing RNA-Seq Workflow Methodologies Using Shannon Entropy.. 2024.
• Commentary: a review of technical considerations for planning an RNA-Sequencing experiment.. 2025.
• From bench to bytes: a practical guide to RNA sequencing data analysis.. 2025.
• Formation of the Vasculogenic Mimicry Phenotype in Melanoma Mel Z Cells Is Coupled with Changes in Inter-Chromosomal Contacts of Developmental Genes with rDNA Clusters.. 2025.
• A comparison of per sample global scaling and per gene normalization methods for differential expression analysis of RNA-seq data. PLoS ONE, 2017.
• In Papyro Comparison of TMM (edgeR), RLE (DESeq2), and MRN Normalization Methods for a Simple Two-Conditions-Without-Replicates RNA-Seq Experimental Design. Frontiers in Genetics, 2016.
• A comparison of RNA-Seq data preprocessing pipelines for transcriptomic predictions across independent studies. BMC Bioinformatics, 2024.
• Erratum: Comparison of Normalization Methods for Analysis of TempO-Seq Targeted RNA Sequencing Data. Frontiers in Genetics, 2020.
• Comparing the normalization methods for the differential analysis of Illumina high-throughput RNA-Seq data. BMC Bioinformatics, 2015.
• Choice of library size normalization and statistical methods for differential gene expression analysis in balanced two-group comparisons for RNA-seq studies. BMC Genomics, 2020.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.