From transcriptomic data to research-ready evidence.
Paired-end bulk RNA-seq in. Cohort quality, differential expression, pathway context, candidate biomarkers and a reproducible report out. For research use only.
Illustrative example built from synthetic data. Not a result from any project.
Follow a run, stage by stage
Six stages, from sequencing files to a reproducible report. Select a stage to see the question it answers and the kind of output it produces.
Illustrative example built from synthetic data. Not a result from any project, and selecting anything here does not run an analysis.
Paired-end RNA-seq, with sample groups
A run starts from a folder of paired-end bulk RNA-seq files and a table naming each sample and the group it belongs to. The supported question is tumour versus normal in human colon tissue.
Which samples are fit to analyse
Reads are trimmed, aligned and counted per gene. Each sample is then checked on how well it mapped and marked pass, warn or fail. Failing samples are excluded and named in the report, and a run stops rather than analysing a cohort too small to support a comparison.
Which genes differ between the groups
Differential expression compares the groups gene by gene, reporting the size of the change and an adjusted statistic for each. Select a highlighted point to read how one is described.
What biology those genes sit in
Candidate genes are tested against public annotation sets to see which biological themes contain more of them than chance would give. Enrichment generates hypotheses; it does not establish causality.
Which genes survive every rule
Candidates must hold up across the selection rules, including a robustness check that refits the comparison with each sample left out in turn. Genes that only pass in one arrangement of the data do not become candidates.
What a run hands back
A run produces one report with its tables, figures and quality summaries, alongside a record of the parameters, tool versions and checksums needed to repeat it.
Select another stage to follow the workflow.
Find the candidates, or score against a model that exists.
Every run shares the same route up to the point where the data is counted and checked. What differs is the question asked of it afterwards.
- Input and metadata checks
- Trimming
- Alignment and counting
- Sample quality: pass, warn or fail
- Cohort quality
A run can stop here on request, so the quality of the cohort can be read before any analysis.
Find candidates in your cohort
For a cohort where you do not yet know which genes separate the groups. The comparison is run, placed in pathway context, and narrowed to the genes that survive every selection rule.
- Expression tables and plots
- Pathway results
- A candidate table with the direction of change
- A report and run record
If too few genes survive the rules, no model is fitted and the report says so. That is not a negative finding about your samples.
Score samples with a locked model
For the one question a fixed model already exists for. The model is applied as it stands: nothing is refitted, no threshold is tuned to your data, and its files are checked against their hashes at the start and end of every run.
- A score for each sample
- An evaluation table
- A report and run record
A locked model is valid only for the question and the counting route it was built on. Independent validation of our own locked model has not yet been carried out.
A result you can return to
Every run records what it needs for the same result to be reproduced, or audited, months later.
Every parameter
The full launch configuration of the run, recorded as it was actually used rather than as it was meant to be.
Every tool version
The installed version of each tool in the run, so a later run can be compared with this one on equal terms.
Checksums
Hashes of the inputs, the configuration and each report file, so any later change to them is detectable.
What was excluded
Any sample dropped at quality control is named with the reason, rather than quietly disappearing from the cohort.
For research use only. Not a clinical or diagnostic device. Candidate findings require appropriate independent validation. Candidates come from the cohort you supply; small cohorts give unstable results, and a candidate set is only as general as the data behind it. Performance measured on the samples that built a model is not validation, and independent validation of our own locked model has not yet been carried out. The supported question is tumour versus normal in human colon tissue; prognostic and treatment-response questions are not implemented.
Have RNA-seq data and a question to put to it?
Tell us about your study and the comparison you want to make, and we can discuss whether this workflow suits it and where it would start.