Scan a sequence and watch hits appear as you type, search a motif against reference databases, see where it sits among every known motif, or match a whole set at once. The analysis you already run — now with nothing to install, no job to wait on, and nothing uploaded.
All four run in your browser on the same motif databases — JASPAR, HOCOMOCO, CIS-BP, and Vierstra for DNA plus CIS-BP-RNA — take your own MEME or JSON files, and build to single HTML files you can keep.
Paste a DNA or RNA sequence and see every motif occurrence highlighted in place, scored the way FIMO scores them.
Type a consensus (IUPAC codes welcome) or upload a matrix, and rank the closest database motifs by three complementary measures.
Every motif in every bundled DNA database laid out by similarity, coloured and labelled by motif family — and your own motif or sequence placed on it.
Compare a whole set of motifs — or the filters a trained model learned — against the databases, clustered and aligned so you can see what they resemble.
A hit is only useful if you know what it means. The atlas gives every one of the 4,232 bundled DNA motifs a curated family, and every family a short referenced entry: what binds the site, what the sequence means structurally, and what a match does and does not let you conclude.
Grouped by what the motif is, not by gene name — a family here is a set of matrices it is reasonable to treat as the same motif, usually one DNA-binding-domain family whose paralogues cannot be told apart from sequence.
The factor name on a motif can be wrong: ChIP-derived matrices can belong to a partner protein, 83 CIS-BP entries are inherited from a non-human orthologue, and archetype names describe a majority rather than a protein. Each entry says which applies.
Every motif is compared with all 4,232 others, so the atlas can flag the cases where a motif's closest relatives disagree with the name it carries — and where an uncharacterised zinc finger is simply reproducing a known family's motif.
The tools implement well-established motif methods in TypeScript so they run in the browser, and validate their output against reference implementations where one exists.
A probability matrix is converted to a log-odds PWM against a background distribution (default A/C/G/T = 0.295 / 0.205 / 0.205 / 0.295) with a small pseudocount. A window's score is the sum of per-position log-odds.
The full score distribution under the background is built by dynamic-programming convolution over motif positions, quantized to a fixed bin size. Each hit's p-value is the exact tail probability of its score — no sampling, no normal approximation.
The query and each database motif become columns of log-odds scores. Sliding the query across a motif, every offset is scored by normalized cross-correlation — the cosine similarity of the overlapping columns, from −1 to +1 — and the best offset is kept over both strands. A complementary log-likelihood ratio sums the query's log-odds over that overlap. It's a simple, sensible similarity measure; an FFT makes the all-offsets comparison fast.
Tomtom mode runs the MEME-style column-comparison statistic with complete (exact) p-values, giving calibrated significance for motif-vs-motif matches. Ported from memesuite-lite and checked for parity against its test vectors.
Bundled and selectable in both tools. Please cite the original source for any database you use.
Select any combination of DNA databases to search them together; RNA is scanned on its own. Bring your own too — upload any MEME-format file (ACGT or ACGU) or a JSON matrix, parsed in your browser and never uploaded.
Each tool builds to one self-contained file — databases embedded, no CDN, no server. Save it and open it from disk; it works fully offline.
All scanning and searching runs in your browser. Sequences you paste are never sent anywhere.
The one exception: if you enter genomic coordinates in the scanner (Seqstr format), those coordinates — not sequences you typed — are sent to the UCSC Genome Browser API to fetch the reference sequence.