Working with Your Own Data

Whatever pipeline produced your data (a core facility, a GEO download, your own alignment), bulk RNA-seq analysis starts from the same pair of tables. The first holds the measurements. The second holds the experimental design. Sample names are what join them.

The count matrix has one row per gene and one column per sample. Each cell is the number of sequencing reads assigned to that gene in that sample. The sample sheet (often called metadata or colData) flips the orientation: one row per sample, with columns describing each one, such as which condition it belongs to, which batch it was processed in, and anything else you recorded.

TransXplorer reads the count matrix with a few simple assumptions. If your file follows them, it loads first time. The same rules appear in the app under File Format Guide & Tips in the upload sidebar.

• Header rowrow 1 The first row holds column names: a label for the gene column, then one name per sample. No title rows, notes or blank lines above it.
• Gene ID columngene_id TransXplorer looks for a column named like gene_id, GeneID, Gene, Symbol, gene_symbol, GeneName or ID (any capitalisation). If none is found it uses the first column. Putting IDs first is the safe choice. Rows with an empty ID are dropped.
• Sample columnsnumeric only Every column other than the gene ID column is treated as a sample. That includes columns you might not think of as samples, such as Length, Chr or a description column, so delete those first.
• At least two samples≥ 2 columns The upload is rejected with fewer than two numeric sample columns. For a meaningful comparison you need far more than that; see pitfall 5.
• Gene levelone row per gene Upload gene-level counts. If your quantifier reports transcripts (Salmon, kallisto, RSEM isoforms), summarise to genes first, for example with tximport. The FASTQ Processing tab in TransXplorer does this step for you.

Accepted file types

The Upload Count Matrix (CSV/XLSX/TSV/TXT) button takes four formats. Files up to 500 MB are accepted.

Which gene IDs work

Use whatever your pipeline produced, as long as the whole file uses one type. The identifiers are kept as row names for the differential expression step. Later, when modules such as enrichment or network analysis need gene symbols, TransXplorer converts Ensembl and Entrez IDs to symbols using the organism's annotation database.

The same gene in the same sample can be reported three different ways depending on where you got the file. Only one of them is what differential expression software expects.

Why DE tools insist on counts

DESeq2 and edgeR model each gene's reads with a negative binomial distribution, whose variance depends on how many reads were counted (Love et al., 2014; Robinson et al., 2010). A gene with 3 reads is far noisier than one with 3,000, and the model uses that difference to decide how much to trust each fold change. limma-voom estimates the same mean–variance relationship from counts. Once you convert to TPM or FPKM, the information about how many reads were actually observed is gone: 3 reads and 3,000 reads can end up as similar-looking TPM values. These tools also do their own normalisation for library size, so feeding them pre-normalised values normalises twice. Wagner et al. (2012) showed that even comparing RPKM values across samples is unreliable, which is one reason TPM was proposed.

Where to find the raw counts

• featureCounts / HTSeq-countinteger counts Already raw counts. Drop the annotation columns from featureCounts and the __no_feature-style summary rows from HTSeq.
• STAR --quantMode GeneCountsReadsPerGene.out.tab Pick the column that matches your library strandedness, and remove the four N_ summary rows at the top.
• Salmon / kallisto / RSEMestimated counts Use NumReads (Salmon), est_counts (kallisto) or expected_count (RSEM), summarised to genes. These have decimals but are on the count scale, which is fine (Soneson et al., 2015). Don't use the TPM column.
• GEO supplementary filescheck the label Files labelled raw counts or read counts are what you want. Files labelled normalised, FPKM or TPM are not. TransXplorer's Or fetch from GEO box can pull NCBI-processed counts for many RNA-seq series directly; see the GEO tutorial.

The count matrix says how much of each gene there is. The sample sheet says what was done to each sample. Without it, software can only guess which columns to compare.

Keep it small and plain. A minimal sample sheet that works everywhere in TransXplorer looks like this:

Sample,Group,Batch
Ctrl_1,Control,A
Ctrl_2,Control,B
Ctrl_3,Control,C
KO_1,KO,A
KO_2,KO,B
KO_3,KO,C

• Samplerequired The key column. Name it exactly Sample. In the group dialog TransXplorer can also recognise headers such as sample_id or SampleName, but the batch-metadata upload insists on Sample, so using that name everywhere avoids surprises. Values must match the count-matrix column headers.
• Groupthe comparison The condition each sample belongs to: Control/KO, Healthy/Disease, time points, and so on. A column named Group, group or Condition is pre-selected as the group column for you.
• Batchoptional Sequencing run, library-prep day, plate, donor: anything technical that could shift expression. A column named Batch or batch is pre-selected as the batch column.
• Other covariatesoptional Sex, age, RIN, and similar. These can be chosen under Additional Covariates when you configure the metadata.

Exact match, or no match

Sample names are compared as plain text. TransXplorer doesn't guess that ko_1 probably means KO_1. Each mismatch below would leave a sample unassigned:

No metadata file? Let the names carry the groups

If you skip the sample sheet, TransXplorer tries to read groups from the sample names. It takes the text after the last underscore as the group, so S1_Control becomes Control. It accepts the result only if there are at least two groups and no more groups than half the number of samples. The app's own format hint shows the same pattern: Sample1_Control.

1. 1

   Load the count matrix

   Transcriptome Analysis › 1. Data Input & Setup

   Click Upload Count Matrix (CSV/XLSX/TSV/TXT) and choose your file. TransXplorer checks the file is not empty and under 500 MB, reads it, finds the gene ID column, and confirms there are at least two numeric sample columns. If something is wrong you get a message naming the problem, for example that no numeric expression data was found.

   The same panel offers two shortcuts: Or fetch from GEO (type a GSE accession and click Fetch) and Load Example Counts, which loads the GSE151427 dataset. Loading the example once is a good way to see what a correctly formatted file produces.

2. 2

   Confirm the sample groups

   Dialog › Sample Group Assignment

   A dialog opens as soon as the counts load. You have three ways to define groups:

   • Auto-detected. If the sample names follow the Sample_Group pattern (fig 4), you'll see the proposed groups with a summary. Click Accept These Groups, or Choose Different Method if they're wrong.
   • Manual Assignment. Type group names, click Add Group, then pick a group for each sample from a dropdown.
   • Upload Metadata. Choose your sample sheet (CSV, TSV, TXT or XLSX), then pick which column holds the groups. TransXplorer shows how many samples matched and how many fall in each group.

   Finish with Confirm Groups.

3. 3

   Select the organism

   Sidebar › Select Organism

   The Select Organism: * menu lists Human, Mouse, Rat, Drosophila, Zebrafish, C. elegans, Yeast, Arabidopsis, Chicken, Pig and Cow. Anything else goes under Other, where you can type the organism name plus an optional KEGG organism code and NCBI taxonomy ID. As the app notes, differential expression, QC and WGCNA work for any organism; pathway enrichment and protein networks need the KEGG code or taxonomy ID; drug targets and cell-type deconvolution are human-only.

4. 4

   Decide how to handle batch

   Sidebar › Batch Effects & Covariates Detection

   Once groups are confirmed, a batch panel appears with three options: Automatic Detection (recommended; scores every metadata column with PVCA, silhouette and kBET), Manual Selection (you name the batch column), or Skip.

   For Automatic or Manual, use Upload Batch Metadata File. This file must contain a column headed Sample that matches your count-matrix names. It can be .csv, .tsv, .txt or .xlsx; tab, semicolon or comma separators are detected automatically. A Configure Sample Metadata window then asks for the Group Column (Required), an optional Batch Column, and any Additional Covariates. Your sample sheet from Section 4 can be reused here unchanged.

5. 5

   Set parameters and run

   Sidebar › 2. Analysis Parameters

   Choose the DEG package (edgeR, DESeq2 or limma-voom), the normalisation method and the comparison, then set thresholds. Click Run Exploratory QC first to look at PCA and sample clustering, then Run DEG Analysis. If any group has fewer than two samples, TransXplorer shows a warning that statistical power may be low. For a guided tour of the results screens, see Your First RNA-Seq Analysis.

Tick each item as you check your files. Nothing is saved or sent anywhere; the boxes reset when you reload the page. It prints cleanly if you'd like a paper copy at the bench.

1. Ziemann M, Eren Y, El-Osta A. Gene name errors are widespread in the scientific literature. Genome Biology. 2016;17:177. doi:10.1186/s13059-016-1044-7
2. Abeysooriya M, Soria M, Kasu MS, Ziemann M. Gene name errors: lessons not learned. PLoS Computational Biology. 2021;17(7):e1008984. doi:10.1371/journal.pcbi.1008984
3. Bruford EA, Braschi B, Denny P, Jones TEM, Seal RL, Tweedie S. Guidelines for human gene nomenclature. Nature Genetics. 2020;52:754–758. doi:10.1038/s41588-020-0669-3
4. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology. 2014;15:550. doi:10.1186/s13059-014-0550-8
5. Robinson MD, McCarthy DJ, Smyth GK. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics. 2010;26(1):139–140. doi:10.1093/bioinformatics/btp616
6. Wagner GP, Kin K, Lynch VJ. Measurement of mRNA abundance using RNA-seq data: RPKM measure is inconsistent among samples. Theory in Biosciences. 2012;131:281–285. doi:10.1007/s12064-012-0162-3
7. Soneson C, Love MI, Robinson MD. Differential analyses for RNA-seq: transcript-level estimates improve gene-level inferences. F1000Research. 2015;4:1521. doi:10.12688/f1000research.7563.1
8. Schurch NJ, Schofield P, Gierliński M, et al. How many biological replicates are needed in an RNA-seq experiment and which differential expression tool should you use? RNA. 2016;22(6):839–851. doi:10.1261/rna.053959.115