Examples

Curated, read-only examples built from real stored computations. They are computational results and hypotheses, not experimental findings or validations. Listed from a static catalogue, so this page works with the backend stopped.

Example 1

Real computed result (GNINA run)

EGFR (PDB 1M17) with erlotinib - redocking control

Small-molecule docking · GNINA 1.3.3 (Kaggle P100)

Erlotinib redocked into its own 1M17 chain A co-crystal site: 9 poses, 125 interactions, 9 reference RMSDs. RMSD to the crystal ligand: top-ranked pose (pose 1) 1.906 A; lowest of the 9 reported poses 1.20 A (pose 2, ranked second).

Computational hypothesis; a redocking control, not a binding validation.

Open Example 1

Example 2

Real computed result with caveats (GNINA run)

TRPV1 (PDB 11CO) with 2-[(2,3-dimethylphenyl)amino]benzoic acid

Small-molecule docking · GNINA 1.3.3 (Kaggle P100)

A public molecule docked into the site of the co-crystal ligand of TRPV1 (chain A of a tetramer, cryo-EM 2.49 A): 5 poses, 43 interactions.

Hypothesis demonstration, not a control. The binding-site box is shown as RESEARCH_HYPOTHESIS (a search region); its stored derivation is DERIVED_RESULT - different objects, the conservative label is displayed. Pose 1 (-7.29) is well separated from the rest (-4.1 to -4.8); GNINA logged 'initial pose not within box' for some starts.

Open Example 2

Example 3

Real computed result (platform ADMET service)

Erlotinib - ADMET Tier-1 descriptor profile

ADMET Tier-1 (RDKit descriptors) · RDKit 2026.03.6

Seven Tier-1 endpoints computed by the platform ADMET service on the pinned RDKit.

Computed descriptors, not ADME predictions; applicability NOT_EVALUATED; no composite score. Generated 2026-10-03 on local SQLite, not a historical production run.

Open Example 3

Example 4

Real computed result (platform generative service)

Erlotinib-seeded RDKit enumeration with PAINS screen and ranking

Deterministic enumeration + ranking · rdkit_enumeration 2026.03.6

BRICS recombination (2 candidates) and scaffold decoration (8 candidates) of erlotinib; ranked with validity, novelty and rule-of-five; PAINS catalogues A/B/C screened.

Enumeration, not de-novo design. Constraint and synthetic components are unavailable and left out, not guessed. Candidates are computational hypotheses. Generated 2026-10-03 on local SQLite.

Open Example 4

Example 5

COMPUTED RESULT - real Boltz-2 execution on a GCP L4

EGFR (PDB 1M17) with erlotinib - Boltz-2 co-folding canary

Structure + affinity co-folding (benchmark view) · Boltz-2 2.2.1 (NVIDIA L4, GCP)

One real Boltz-2 run: single-sequence co-fold of 1M17 chain A + erlotinib. Engine-native outputs only: confidence_score 0.937, iptm 0.984, complex_plddt 0.926 (0-1 scale), affinity_probability_binary 0.756, affinity_pred_value -0.890 (Boltz log10 IC50 uM model scale). Structural/reference agreement 3.92 A (geometry only).

Boltz-2 metrics are model predictions on Boltz's own scales; they are not GNINA scores, not experimental affinities and not validation. Single-sequence mode (no MSA) reduces accuracy. The submitted chain is continuous across a 12-residue crystallographic gap, which inflates the global receptor C-alpha deviation. One run, one sample.

Open Example 5