Structure Prediction

Predict, visualize and analyze protein structures with Nilus Fold (powered by Boltz-2.1), then inspect confidence, interfaces and curated annotations.

Protein sequence0 aa

UniProt / gene lookuphuman Swiss-Prot

Target receptor / proteinChain A (FASTA)

Small-molecule ligandChain B (SMILES)

Bioactive receptor presetsBoltz co-modeling

Configures Chain A (Receptor) + Chain B (Ligand) with 1-click binding constraints.

Complex entitiesprotein · DNA · RNA · ligand

Mutatione.g. R274W

A mutation changes the input sequence only. It does not imply pathogenicity or functional impact.

Nilus FoldPOWERED BY BOLTZ-2.1 · COMPLEX CONFIDENCE

●

Prediction typeauto-builds entities

Advanced Nilus Fold controls

SamplesBoltz: 1–10

Protein MSAall protein chains

Recycling stepsdefault 3

Sampling stepsdefault 200 · min 50

Step scale1.3–2.0 · default 1.638

Binding metricoptional

Explicit binder chainnever inferred silently

Expert MSA / template / structural constraints

Custom MSA formatsingle protein only

Custom MSA URLHTTPS

Template guidance · optional · up to 4 supported by Boltz; this preview exposes one explicit template mapping.

Template PDB/mmCIF URLHTTPS

Input chainNilus chain

Template chainsource chain

Force threshold Åoptional

Polymer contact constraint · optional · residue indices are 0-based exactly as required by Boltz.

Token 1 chain

Token 1 residue0-based

Token 2 chain

Token 2 residue0-based

Maximum distance Å

Pocket constraint · optional · binder chain plus target residue set.

Binder chain

Target chain

Target residues0-based comma list

Maximum distance Å

Force pocket constraint

Use constraints only when they are justified by prior structural/experimental knowledge. Nilus records them in provenance; constrained predictions must not be presented as unconstrained evidence.

Interpretation: more samples explore stochastic alternatives; they do not guarantee a correct structure. Automatic MSA is the default. Single-sequence mode deliberately removes MSA information. These controls map directly to Boltz-2.1 inputs.

Research-use computational prediction. Confidence scores and geometric contacts are not experimental validation, binding affinity or clinical evidence.

Backend readyEstimate before run

Recent predictions

SIRT1_HUMAN

Experimental reference

Human SIRT1 catalytic-domain reference · PDB 4I5I · Boltz metrics not yet available

Interactive 3D Structure

Drag to rotate · Scroll to zoom · click residues to inspect

pLDDT confidence

Very high (90–100)

High (70–90)

Low (50–70)

Very low (<50)

Y ↑
Z ↙ X →

Experimental reference preview · PDB 4I5I

Loading scientific reference…Nilus Fold output will replace this preview after a prediction.

Residue confidence map

Very highHighLowVery low

Nilus Fold per-residue confidence is shown only when the returned structure contains confidence values.

PAE matrix

PAE is never simulated.
Nilus displays it only when a real Nilus Fold archive exposes a parseable PAE array.

Key structural features

↯Waiting for predicted structure

✓Experimental SIRT1 reference loaded for preview

Model confidence

—/100 · confidence index

Awaiting Nilus Fold prediction

The primary ring shows Nilus Fold structure confidence directly. pLDDT, pTM, ipTM and PAE remain separate because they measure different uncertainty.

—Complex pLDDT

—Structure conf.

—pTM

—ipTM

Residue confidence distribution

————

Model details

ModelExperimental PDB

InputSIRT1_HUMAN

Length—

Chains—

Job ID—

Runtime—

Completed—

StatusReference

Scientific intelligence suite

P1 evidenceP2 comparisonP3 interfacesP4 preliminary QCP5 ensemble

Each phase keeps evidence provenance explicit. Nilus never merges model confidence, coordinate-derived geometry and experimental evidence into a fabricated accuracy score.

Sample ensemble

—Samples returned

—Structure conf. range

—ipTM range

Nilus Fold can return multiple stochastic samples. Nilus keeps each sample separate instead of averaging coordinates or turning agreement into a fabricated accuracy score.

AlphaFold DB reference

—AFDB mean pLDDT

—Very-high fraction

—AFDB version

Load a reviewed UniProt entry to query the official AlphaFold DB API. AFDB is an independent computational reference, not ground truth for a Boltz prediction.

Residue intelligence

Select any residue in the 3D viewer to inspect confidence, secondary structure, local geometry, curated annotation overlap and neighboring residues.

Coordinate quality audit

—Protein backbone complete

—Potential chain breaks

—Gross atom overlaps (<1 Å)

—Residues pLDDT <70

Coordinate checks are geometry heuristics, not experimental structure validation.

Structural domains / annotations

Run UniProt lookup to load curated features

Domain names are never invented. Known annotations come from UniProt; geometry-derived segments are labeled as predicted regions.

Uncertainty hotspots

Region

Residues

pLDDT

Interpretation

—

Awaiting prediction

Model provenance

Engine—

Samples—

Run hash—

Job—

Completed—

SourceReference

The run hash is calculated locally from the normalized molecular input plus inference controls, binding request and model identifier. Restoring input never starts a paid prediction automatically.

Structure Intelligence Agent

In-situ biophysical reasoning engine

Interactive agent connected to active 3D coordinates, PAE tensors, and binding interfaces. Use prompts below or type any structural query to audit pockets, clashes, and export robotic protocols.

Zero-hallucinationActive 3D CoordinatesOT-2 Robotics API