scroll = zoom · drag = pan · click a point = open · hover = info
Atlas assistant
Filters & metrics
Every fold in the curated distillation set (7,757) is a redundancy-reduced cluster
representative that is novel vs the v341 PDB training set. All filtering happens live in your browser;
settings persist across reloads.
Identity & size
Search
Matches the seq_id, the human-readable name, or the sublibrary.
Length
Design-region length in nucleotides (excludes the fixed primer/barcode padding).
Source / letter
Pick datasets in the header Source menu — check any combination and the table merges them. Under Ribonanza-2 curated, per-library letter checkboxes appear: A–E = synthetic designs (gRNAde, UW, RNAMake), F–H = natural RNAs (RNAcentral). Ribo-1 pseudolabel / OpenKnot / RFAM-PDB are separate sources (no SHAPE/motif columns).
Confidence
pLDDT
Model confidence, 0–100; higher = more confident. The pool bar is pLDDT>85.
clashscore
Steric clashes per 1000 atoms (Phenix); lower = cleaner geometry. Populated on curated reps.
Novelty
best_tm1
USalign TM1 to the closest of the 2,243 v341 PDB-RNA chains. Lower = more novel (≲0.45 novel, ≲0.40 clearly novel). Fold-level for the whole set; the deck candidates use a per-structure value.
nearest
The closest known PDB chain found for that TM1; hover (or see the deep view) for the RCSB entry title — handy for sanity checks.
overlap_AE
For an F–H fold, the TM1 to its closest A–E fold — i.e. how structurally redundant it is across the natural/synthetic sets. Lower = more distinct.
Chemical mapping (SHAPE/DMS)
SHAPE
yes = the fold's tertiary-motif residues are protected (mean 2A3 protection > 0) or the fold's reactivity agrees with its predicted pairing (r(2A3) < −0.2). no = reactivity exists but doesn't indicate protection. n/d = no usable protection signal (no tertiary motif to score, or no reactivity coverage) — the raw reactivity may still be viewable in the fold's deep view. A–E reactivity is read from the experiment HDF5s; F–H from the full design-aligned chemmap.
SHAPE agr
SHAPE–pairing agreement: correlation of 2A3 reactivity with the unpaired positions. Positive = good (reactive where unpaired, protected where paired ⇒ the chemical mapping supports the predicted fold). This is the sign-flipped r(2A3,is-paired).
In a fold's deep view, the pairing track (white = paired, light red = unpaired) sits under the DMS/2A3 reactivity rows so you can visually check agreement: unpaired (light red) should line up with reactive (red) reactivity. Pairing comes from the fold representative's predicted secondary structure.
Structure
Compactness
C1′–C1′ contact ratio: number of nucleotide pairs whose C1′ atoms are within 8 Å (sequence separation ≥ 6), divided by length. A globularity proxy (the RNA analog of the AlphaFold-DB Cα contact ratio). Higher = more compact / globular; an extended chain scores near 0.
Base-paired fraction
Fraction of positions that are paired in the fold's predicted secondary structure (dot-bracket). Higher = more structured. Computed on the fold representative.
Motifs & topology
Motifs
Tertiary RNA motifs detected by Rosetta rna_motif (A-minor, TL-receptor, T-loop, U-turn, platform, …). Filter by "≥1 tertiary", "a rare tertiary", or specific types. tert/rare columns count them.
Tertiary complexity (crossed-pairs)
Fraction of residues in a ‘crossed’ contact pair (Max/Rhiju): a 3D contact pair (i,j) is crossed when another contact (k,l) interleaves it (i<k<j<l) — the residue is pinned by tertiary structure, not just nested 2D pairing. bpp-free 3D topology. Distinct from compactness (density) and a continuous 3D-contact generalization of pseudoknot: it separates genuinely tangled tertiary folds from merely compact hairpins. Threshold contact < 12 Å, |i−j| ≥ 8 (Rhiju to finalize). The “tertiary crossed-pairs” 3D coloring mode paints these pinned residues.
MOHCA-regime fraction
Fraction of predicted contacts in MOHCA-seq’s strongest range (medium-range, 25–50 nt separation) — a proxy for where a MOHCA map would most add signal.
Pseudoknot
The predicted secondary structure contains crossed base pairs.
Ranking & display
Ranking · top-N · per letter
Sort key (novelty, compactness, base-paired fraction, pLDDT, length, …) and how many rows to show — overall, or top-N within each library. Click any column header to sort by it.
Base colors
Toggle the sequence-strip nucleotide palette (A gold · C green · G red · U blue).
Deep view (click any row)
3D structure
Interactive cartoon; tertiary-motif residues shown as colored sticks. The 3D structure coloring dropdown (Display panel) selects how the cartoon is colored: 2A3 / DMS reactivity (blue = protected → red = reactive), pLDDT confidence (blue high → orange low), base pairing (white = paired, light red = unpaired), nucleotide (A/C/G/U palette), or spectrum (5′→3′). Reactivity falls back to spectrum when a fold has no chemical-mapping data. Dock it centered / right / bottom via the header layout buttons.
Tracks
Motif lanes · sequence · DMS · 2A3 reactivity, aligned on the design-sequence position axis.
Atlas assistant