Preserve fixed positions
Family-specific 40S, 60S, and P-site annotations define bases that cannot change during proposal.
Internal ribosome entry site (IRES) · one-dimensional, two-dimensional, and three-dimensional evidence
Scientific Design Engine · Regulatory ribonucleic acid (RNA) design
TrioIRESSearch IRES sequence space while preserving structural intent
TrioIRES designs family-specific intergenic-region internal ribosome entry site RNA under fixed recognition constraints. A candidate moves through activity-model, topology, and selective three-dimensional geometry evidence before receiving a final ranking.
Three experimental structures define the active geometry context
The publication-ready PyMOL gallery shows the resolved IRES chains extracted from ribosome-complex structures for CrPV, IAPV, and TSV.
Evidence boundary. The panels are experimental reference coordinates; they do not show generated candidates or activity.- Design object
- Single-chain IGR IRES RNA
- Active families
- CrPV · TSV · IAPV
- Evidence stack
- 1D · 2D · 3D
- Evidence state
- Reference and computational
01 / Abstract and project definition
Structure-Aware RNA Design
TrioIRES is a computational design system for Class IV intergenic-region internal ribosome entry site (IGR IRES) ribonucleic acid (RNA). It turns a family-specific reference package into constrained sequence proposals, explicit gate decisions, component evidence, and a stable candidate ranking.
The reference package contains sequence length, fixed recognition positions, a literature-consensus two-dimensional partner map, a resolved Protein Data Bank (PDB) chain, and optional recognition groups for the 40S small ribosomal subunit, 60S large ribosomal subunit, and P-site region. The current reference set covers cricket paralysis virus (CrPV), Taura syndrome virus (TSV), and Israeli acute paralysis virus (IAPV). Generated-candidate experimental activity remains a downstream validation endpoint.
02 / Scientific problem
RNA sequence variation is useful only when recognition sites and family topology remain compatible.
The mutable positions form a large constrained search space. A high activity-model score can still accompany broken fixed sites, incompatible topology, or implausible geometry. TrioIRES applies inexpensive sequence and topology gates before selective three-dimensional evaluation, and it keeps the evidence behind each decision separate.
Respect pseudoknot-aware structure
Candidate partner maps are compared with literature-consensus targets and thermodynamic ensemble evidence.
Use resolved structure carefully
Predicted single-chain coordinates are compared with available reference residues while unresolved coverage remains visible.
03 / Method overview
Sequential gates reserve expensive geometry evaluation for candidates that already satisfy sequence and topology evidence.
A constrained tree inserts fixed bases directly and exposes nucleotide choices only at mutable positions. Batched rollout completes candidate sequences. A ten-model encoder ensemble supplies one-dimensional activity evidence; KnotFold and ViennaRNA supply two-dimensional topology and thermodynamic evidence; and one eligible candidate per rollout batch advances to RhoFold+ or OpenFold3 single-chain geometry review.
Activity prior
A ten-model encoder ensemble returns mean activity probability while length and guanine–cytosine content remain hard constraints.
Output · One-dimensional activity evidenceStructural agreement
KnotFold partner maps are compared with conserved and mutable target regions alongside ViennaRNA ensemble defect.
Output · Two-dimensional topology evidenceGeometry evidence
Predicted single-chain coordinates are compared with resolved reference positions, chain connections, local motifs, and recognition groups.
Output · Three-dimensional geometry evidenceDecision separation
A stable final score supports candidate comparison while a transformed search signal serves branch selection and backpropagation.
Output · Ranking plus search guidance04 / Architecture
Every handoff has an explicit scientific contract
Inputs, operations, evidence, and outputs stay named so the source of each decision can be reviewed.
- Reference
Family-specific design object
Sequence length, fixed sites, allowed mutable bases, consensus topology, resolved chain, and recognition annotations.
→ - Proposal
Constrained tree search
Fixed bases are inserted directly; mutable positions expose only allowed nucleotide choices.
→ - Gate 1
Encoder ensemble
Mean activity probability plus length and composition constraints.
→ - Gate 2
KnotFold and ViennaRNA
Pseudoknot-aware partner agreement, conserved-region fit, and thermodynamic evidence.
→ - Gate 3
Selective geometry
RhoFold+ or OpenFold3 single-chain prediction compared with the resolved reference chain.
→ - Output
Candidate record
Sequence, gate state, component readouts, final ranking, search guidance, and provenance.
05 / End-to-end workflow
From scientific input to an evidence-bearing decision
Each stage consumes a bounded object and leaves a reviewable artifact for the next stage.
- 01Reference package
Specify
Assemble one family-specific target with sequence length, fixed sites, two-dimensional topology, resolved chain, and recognition groups.
- 02Search state
Initialize
Create a constrained tree state that inserts fixed bases and exposes allowed nucleotide choices at mutable positions.
- 03Valid branches
Expand
Generate explicit constrained blocks and reject invalid length or fixed-position changes before scoring.
- 04Candidate batch
Roll out
Complete batched suffix candidates to increase the chance that at least one survives inexpensive screens.
- 05Evidence stack
Gate
Apply activity and topology screens, then advance one eligible candidate per batch to geometry evaluation.
- 06Candidate record
Rank
Return component values, gate states, a stable ten-point final ranking, and separate search guidance.
06 / Experimental design
Questions, protocols, and comparison boundaries remain attached
Every study below describes what was varied, what was measured, and where interpretation must stop.
Cricket paralysis virus / PDB 2NOQ
The Class 1 target has complete resolved-chain coverage. Its ViennaRNA partner set is contained within the pseudoknot-aware KnotFold map, which contributes additional pairs.
- Sequence length
- 190 nt
- Fixed positions
- 68
- Consensus pairs
- 61
- Resolved residues
- 190 / 190
Taura syndrome virus / PDB 3J6Y
The Class 2 target exposes model choice: only 18 reference-sequence pairs are shared by the KnotFold and ViennaRNA maps.
- Sequence length
- 201 nt
- Fixed positions
- 88
- Consensus pairs
- 47
- Resolved residues
- 198 / 201
Israeli acute paralysis virus / PDB 6P4G
The Class 2 target has partial coordinate coverage. Fifty positions lack resolved reference coordinates, so geometry interpretation remains coverage-aware.
- Sequence length
- 253 nt
- Fixed positions
- 84
- Consensus pairs
- 57
- Resolved residues
- 203 / 253
07 / Results and evidence
Read the signal with its evidence boundary
Quantitative summaries, structural views, and scientific interpretation remain separate layers.
Active reference families
CrPV, TSV, and IAPV define the current search targets
Activity encoders
Ensemble mean supplies the one-dimensional prior
Geometry evaluation per eligible batch
Selective three-dimensional review follows cheaper gates
Generated-candidate activity claims
Wet-lab activity remains pending
The current public result is a structural reference system
CrPV, TSV, and IAPV provide literature-consensus secondary structures, fixed recognition annotations, and resolved chains that define target-specific design constraints. The resolved-coordinate coverage is 100%, 98.5%, and 80.2% respectively.
Boundary. Reference availability supports target definition and computational comparison; it does not establish activity for generated sequences.Topology-model agreement varies by family
For CrPV, the 42 ViennaRNA pairs are shared with the 61-pair KnotFold map. TSV shares only 18 pairs between its 47-pair KnotFold and 59-pair ViennaRNA maps. IAPV shares 42 pairs while KnotFold contributes 15 additional pairs.
Boundary. Partner-map agreement is structural-model evidence and does not measure translation activity.Candidate-level results remain intentionally pending
The implemented stack can produce activity, topology, and selective geometry evidence, yet the current public package contains reference structures and method evidence. Generated-candidate structure exports and wet-lab activity results remain open.
Boundary. No candidate success state is assigned without the corresponding public evidence.
Pseudoknot-aware arcs encode the Class 1 target
The arc map exposes 61 consensus pairs across 190 nucleotides and marks the 68 fixed recognition positions used by constrained search.
Evidence boundary. A target topology map defines structural intent and does not report generated-candidate function.
A longer Class 2 target carries partial resolved-chain coverage
The IAPV reference combines a 253-nucleotide target, 57 consensus pairs, 84 fixed positions, and 203 resolved residues.
Evidence boundary. Unresolved positions limit any geometry comparison and remain explicit in interpretation.Reading key
Scientific abbreviations
- IGR IRES
- Intergenic-region internal ribosome entry site, an RNA element that initiates translation through a structured ribosome-binding mechanism.
- KnotFold
- A pseudoknot-aware RNA secondary-structure prediction system used for candidate partner maps.
- ViennaRNA
- An RNA thermodynamics toolkit used here for ensemble-defect evidence.
- RhoFold+
- A single-chain RNA three-dimensional structure predictor used for selective geometry evaluation.
08 / Limitations and provenance
What the current evidence can establish
The new page excludes the former coordinate-line projections and uses professional molecular rendering for three-dimensional structure evidence.
Activity remains unmeasured
The public package does not contain generated-candidate activity distributions, assay protocols, controls, or completed wet-lab results.
Single-chain geometry is partial context
RhoFold+ and OpenFold3 candidate structures omit the complete ribosome-bound environment and remain model comparisons.
Reference coverage varies
IAPV lacks coordinates for 50 target positions, and the auxiliary Plautia stali intestine virus (PSIV) reference has only 32 of 196 residues resolved.
Topology models can disagree
KnotFold and ViennaRNA partner maps differ substantially for TSV, so model choice remains part of the evidence rather than a hidden assumption.