About · SMARTFlexDB

An end-to-end example

Each entry pairs a ligand-bound (holo) RNA with a ligand-free (apo) structure of a closely matched RNA. Rather than list everything, we walk through one representative pair feature by feature — open it and follow along.

overlay

Example · the TPP riboswitch

The E. coli thiM TPP riboswitch - a widely studied candidate antibacterial target — bound to a small synthetic fragment (4NYB, ligand 2QC) versus its ligand-free aptamer (8F4O). It shows one of the largest conformational changes in the set.

Open this example pair →

Each feature is tagged by where its data comes from: deposited metadata taken verbatim from the PDB entry · tool-derived produced from the deposited coordinates by an established tool (x3dna-dssr, US-align, BLAST, RDKit) · geometry-derived measured directly from the coordinates here (gemmi / NumPy) · heuristic a rule-based inference — treat as a hypothesis.

How pairs are selected

Before any of the per-pair features below are computed, an apo–holo pair has to pass four gates:

Closely matched RNA: ≥90% identity (BLASTN). Each holo RNA chain is searched against candidate apo structures with BLASTN. A hit is kept only if pident ≥ 90% and the alignment covers ≥ 90% of both the query and the subject chain (qcovs ≥ 90 and scovs ≥ 90). So apo and holo are closely matched RNA at ≥90% identity over ≥90% of each chain — there is a hard ≥90% identity floor.
Same pocket. The binding site is the set of RNA residues with any atom within 8 Å of the ligand (a wider 10 Å shell is used for pocket-level clustering). The apo must reproduce that pocket: every pocket position has to map onto the apo sequence and match the same base, and for multi-chain pockets this must hold across all pocket chains. Apo structures that don't present the same pocket are rejected.
Comparable length. Pairs whose compared chains differ in length by more than 20% are dropped as unlikely to be a closely matched RNA. The lengths compared are the declared (SEQRES) lengths the depositors deposited, not the modeled nucleotide counts, so a chain with unmodeled density is not penalized for it. On this release the gate removes nothing. The largest declared-length difference between a holo chain and any apo offered for it - default and alternative apo structures alike — is 8.7%, so this is a forward guard on future input. Note that the bulk download's holo_len / apo_len columns count modeled nucleotides. A served pair can therefore differ by more than 20% there without failing this gate (8OLZ vs 9QU7: 23.8% modeled but 1.3% declared — 9QU7 models 297 of its 396 nt).
RNA-only gallery. The served gallery is restricted to RNA. A pair is hidden if the complex contains any protein or DNA chain (holo or apo side), or if the "apo" is itself ligand-bound in this pocket (not a true apo). It is also hidden if the holo "ligand" is the RNA's own covalent 5′-terminus. The same applies if the holo "ligand" is an ion, crystallization additive, or heavy-atom phasing soak (e.g. selenate, or a gold / iridium ion — not a small molecule). After these gates, the RNA-only pairs are served.

Net effect on dataset bias: a ≥90% apo–holo identity floor, same-pocket and ≤20%-length-difference matching, and an RNA-only restriction — a curated subset served out of the larger underlying set. The Database gallery opens on the full release - both identity facets start on the whole span present in the data, so nothing is hidden until you narrow one. The bulk download always contains the full served set.

The tour — what each feature shows for this pair, and where it comes from

Feature	What you see here	Where it comes from
Basic Information	both structures' method/resolution/authors + ligand properties	RCSB Data API (live) + RDKit deposited metadatageometry-derived
3D superposition + morph	the superposed apo and fragment-bound holo aptamer	sequence-anchored Kabsch on the paired C1′ atoms (a US-align fit is computed alongside and the lower-RMSD one kept; Kabsch is the analytic minimiser of that RMSD, so it is what wins on every served record); morph = interpolation, not MD tool-derivedheuristic
Per-residue displacement	which nucleotides move, and how far (C1′)	sequence-anchored Kabsch (lowest-RMSD transform; see above) tool-derived
Conformational dashboard (feature track)	a per-nucleotide apo→holo C1′ displacement heatmap over the holo sequence (with a x3dna-dssr motif band); radius of gyration and ligand burial are reported as static apo/holo endpoint values in the Overview, not morph curves. A visualization between two endpoint structures, not MD	geometry on the best-of US-align / sequence-anchored Kabsch superposition (burial via biopython SASA) geometry-derived
Feature track (per-nucleotide)	each holo nucleotide as a cell in the feature track: apo→holo displacement color + binding-site marker + DSSR-motif band; click → 3D	geometry on displacement + interactions + DSSR motifs geometry-derived
TM-score & radius of gyration (overall fold)	overall fold similarity (TM-score) and global compaction (R_g, apo → holo) on the Overview. R_g is measured over the apo–holo shared (aligned-core) residues. TM-score is length-sensitive: below ~30 nt it tracks chain length, so read high = fold conserved (not low = fold change) for short RNA	US-align TM-score + gemmi R_g over shared-core C1′ geometry-derived
Sequence alignment (apo vs holo)	the two states lined up residue-by-residue	Needleman–Wunsch geometry-derived
Secondary structure (2D)	base-pairing of apo vs holo; click a base → 3D - what DSSR extracts ↓	x3dna-dssr tool-derived
Ligand–RNA interactions (3D)	H-bonds / contacts to the fragment, with Å labels	gemmi contact geometry on the holo structure; H-bond vs contact typing from x3dna-dssr `--get-hbond` geometry-derivedtool-derived (x3dna-dssr)
Interaction-data tables	pairs · H-bonds · water bridges · base pairs · BP changes · motifs - what DSSR extracts ↓	DSSR (pairs/motifs) + gemmi (contacts) tool-derivedgeometry-derived
Statistics (site-wide)	where this pair sits among all pairs (Δ, resolution, composition)	aggregation geometry-derived

Secondary structure & base pairs — what x3dna-dssr extracts

We run x3dna-dssr on each 3D structure (both apo and holo). Straight from the atomic coordinates — derived from the experimentally determined structure, not predicted from sequence - it reports:

2D layout

the sequence and dot-bracket notation of the folded chain → drives the Secondary structure (2D) diagram.

Base pairs

every pair - canonical and non-canonical - with the bases (e.g. G–U), the Leontis–Westhof geometry type (e.g. cWW, tHS) and the Saenger class → the Base pairing table; the apo–holo set difference becomes Base-pair rewiring.

Structural motifs

stems, hairpin loops, internal loops, bulges, multi-way junctions and single-strand segments (type · residues · size) → the Structural motifs table.

Reading the Leontis–Westhof code (cWW, tHS, …)

Each base has three edges - Watson–Crick, Hoogsteen, and Sugar. A base pair is named by the glycosidic-bond orientation (cis or trans) plus the two interacting edges — 12 geometric families in all. Examples: cWW = cis Watson–Crick/Watson–Crick (the standard A–U, G–C, G–U pairs); tHS = trans Hoogsteen/Sugar-edge (a common non-canonical pair). This geometry is read from the experimental 3D structure, so non-canonical pairs the dot-bracket can't show still appear in the Base pairing table.

Underlying structures come from RCSB PDB; the pair list & pockets from the SMARTFlexDB apo–holo dataset. Everything else is computed locally by x3dna-dssr, US-align, RDKit and gemmi. Note: the morph is a straight-line interpolation between the two endpoint structures (not a molecular-dynamics trajectory). The RNA type is NAKB's curated functional class of the host chain (with a title-keyword fallback for the few structures NAKB does not annotate).

License & data sources

The SMARTFlexDB derived dataset — the apo–holo pair list, geometric descriptors, alignments and cluster assignments — is released under CC‑BY‑4.0. The underlying atomic coordinates are retrieved from the RCSB PDB and remain subject to the PDB's terms of use — no experimental data are redistributed beyond what is already public in the PDB.

SMARTFlexDB is built on these tools: x3dna-DSSR (secondary structure & motifs); US-align (the TM-score, and a comparison fit) and a NumPy Kabsch fit (the served apo–holo superposition / RMSD); BLAST (sequence identity). It also uses MMseqs2 and RM-align (sequence / structure clustering); PLIP (interactions); RDKit (ligand properties); and gemmi (structure handling).

Contact & feedback

Questions, corrections, bug reports, or a candidate apo–holo pair we missed? Email yanjun.li@ufl.edu and jiang.shiyu@ufl.edu - we welcome feedback and respond to data corrections promptly. You can also run the full analysis on your own apo and holo structures from the Query page.