Can I use TPM or normalized values?

No. The model needs raw read counts; it estimates its own size factors, and normalized values break its mean-variance relationship.

Which background should over-representation use?

The genes your experiment tested. Every gene a set annotates is the default of several tools, and it inflates enrichment for sets of genes that are rarely expressed.

Why does a gene have a p-value but no adjusted p?

Independent filtering set it aside for its low mean count, so it was never part of the adjustment: untested, not unchanged. Genes with an outlying count lose their p-value too, and the table says which reason applies.

Is anything uploaded?

No. Every calculation runs in the browser on the files in your project.

Sequencing

RNA-seq: differential expression with DESeq2, over-represented gene sets, gene set enrichment

Find the genes whose expression differs between conditions from raw RNA-seq read counts with the model DESeq2 fits, then ask which gene sets those genes over-represent and which sets sit at the top or bottom of the whole ranking. It reads a count table, a sample table and gene sets in GMT format, and every gene and every set is checked against DESeq2 and enrichit running in R.

Open RNA-seq: differential expression with DESeq2, over-represented gene sets, gene set enrichment Open the RNA-seq: differential expression sample

pasilla knockdown in fly S2 cells, 14,599 genes in seven libraries, for DESeq2 and gene set enrichment against every fly GO term

What it does

01DESeq2 differential expression from raw counts: median-of-ratios size factors, shrunken dispersions, the negative-binomial GLM of any design formula
02Wald tests of a coefficient or contrast, against zero or a fold-change threshold, and likelihood-ratio tests against a reduced design
03Cook's-distance outlier flagging and replacement, independent filtering, and adjusted p-values with the reason any is missing
04Volcano, MA and dispersion plots of every gene
05Over-representation of gene sets by the hypergeometric test against the genes the experiment tested, with fold enrichment, odds ratio, adjusted p and q-values
06Preranked gene set enrichment with normalized scores, multilevel p-values, leading edges and each set's running score

What it runs on

Faithful ports of DESeq2 1.52.0 and enrichit 0.2.6 (the engine of clusterProfiler), checked against the real packages in R over every gene of the pasilla knockdown and every fly Gene Ontology term.

Everything runs client-side. Files are parsed in the browser and projects are saved on this machine, so an unpublished sequence is never uploaded.

Questions

Can I use TPM or normalized values?: No. The model needs raw read counts; it estimates its own size factors, and normalized values break its mean-variance relationship.
Which background should over-representation use?: The genes your experiment tested. Every gene a set annotates is the default of several tools, and it inflates enrichment for sets of genes that are rarely expressed.
Why does a gene have a p-value but no adjusted p?: Independent filtering set it aside for its low mean count, so it was never part of the adjustment: untested, not unchanged. Genes with an outlying count lose their p-value too, and the table says which reason applies.
Is anything uploaded?: No. Every calculation runs in the browser on the files in your project.

Also searched as

DESeq2 online · differential expression analysis · RNA-seq differential expression · volcano plot · MA plot · negative binomial GLM for counts · likelihood ratio test for RNA-seq · gene set enrichment analysis · GSEA preranked · GO over-representation analysis · hypergeometric test for gene sets · clusterProfiler enricher · fgsea multilevel