Tc-43/ORMDL3_Ceramide_Site_Designs_7YIU

ORMDL3–SPT Ceramide-Sensing Site — GA-II Designed Ligand–Receptor Complexes

Why this target matters. The 17q21 locus is the most reproducible genetic association in childhood-onset asthma and acts through ORMDL3 expression — and it drives airway hyperreactivity independently of allergy and inflammation. Every approved asthma biologic targets inflammation, so the patients this mechanism describes are precisely the ones those drugs fail.

477 small molecules generated de novo by the Technetium `TC-43.ai` engine (GA-II), conditioned on the ceramide-sensing regulatory site of the human ORMDL3–serine palmitoyltransferase (SPT) complex. Each molecule was constructed against this pocket rather than selected from a compound library — docking (AutoDock Vina) came afterwards, to place and score the generated molecules in the site. Every design is supplied as a single-file protein–ligand complex in PDB format (471 unique ligand structures).

SPT catalyses the first and rate-limiting step of de novo sphingolipid synthesis and is held under negative feedback by ORMDL3. Ceramide, the pathway's end product, binds the ORMDL3–SPT complex and drives inhibition of SPT. The site targeted here is that regulatory pocket — at the interface between ORMDL3, SPTLC2 and ssSPTa — and not the catalytic site.

Receptor note: coordinates are derived from RCSB 7YIU (human SPT–ORMDL3 complex, cryo-EM 2.9 Å, C6-ceramide bound). Chains A/E SPTLC1, B SPTLC2 (+ PLP cofactor), C ssSPTa, D ORMDL3. The C6-ceramide and a modelled hexadecane were removed so the regulatory site is empty. The same rigid receptor is used in every complex.

Dataset summary

These are de novo generative designs produced by the Technetium GA-II pocket-conditioned generative platform. Each design is docked into the target pocket and scored with AutoDock Vina.

Each complex file is self-contained — receptor structure, the ligand's 3D docked pose, and a 2D↔3D atom map all travel inside the single PDB.

Design objective — read before defining "active"

The therapeutic direction is de-repression. The intent is to displace ceramide so feedback inhibition cannot be applied, increasing SPT flux. A molecule that inhibits SPT would lower sphingolipid synthesis further and is predicted to worsen the disease.

A binder at this site can act in either direction — displacing the feedback signal, or mimicking ceramide and stabilising the inhibited state — and binding affinity cannot distinguish the two. Any functional assay must read SPT enzymatic activity or sphingolipid flux, never binding or ceramide displacement alone.

Pharmacophore, and how this set scores against it

The site is defined by the experimental ligand's own footprint; the pharmacophore transfers ceramide's hydrogen-bond pattern onto non-lipid scaffolds.

HBA1 is the only cross-chain polar contact — engaging it distinguishes a molecule that spans the ORMDL3–SPTLC2 interface from one bound to ORMDL3 alone.

This set was filtered on docking score (floor −12.0 kcal/mol), not on the pharmacophore. Scored against the two required features:

476 of 477 molecules carry a primary carboxamide anchor, but 240 use it as a bidentate clasp on ORMDL3 Asn13 (N–H to OD1, carbonyl folded back to ND2) rather than bridging to SPTLC2. Only 3 reproduce ceramide's single-amide bridge across the interface. Among molecules whose amide N–H sits on Asn13, the carbonyl lands a median 6.73 Å from Tyr122 OH — essentially the Asn13–Tyr122 separation itself (6.74 Å), pointing directly away from SPTLC2.

Docking score does not discriminate: r(Vina, anchored) = −0.09, while r(Vina, heavy atoms) = −0.54. None of the top 10 by score satisfy both required features. Do not rank these molecules on docking score — use the passes_all_gates column in designs.csv.

What holds across the whole set: no molecule approaches the PLP cofactor (minimum 25.9 Å), so the mechanism-reversal risk is clean, and none is a lipid mimic (longest aliphatic chain 6 C). Known liabilities: 96 molecules (20%) exceed cLogP 5, and 197 (41%) are cationic — worth scrutiny in a membrane-embedded site with one wall facing lipid.

Property profile

Computed with RDKit over the 477 designs; docking energy from the generation/docking pipeline.

File format

Each *_cmpx.pdb bundles the receptor and one docked ligand pose:

Two convenience files accompany the complexes:

Key designs.csv columns: passes_all_gates (both required features + cLogP ≤ 5), HBD1_H_to_ASN13_OD1_A, HBA1_to_TYR122_OH_A, primary_amide_binding_mode (bridge_ASN13_to_TYR122 / bidentate_clasp_ASN13 / other), min_dist_to_PLP_A.

Usage

import glob

def read_complex(path):
    smiles, idx = None, {}
    with open(path) as fh:
        for line in fh:
            if line.startswith("REMARK SMILES IDX"):
                toks = line.split("IDX")[1].split()   # flat list of (smiles_pos, atom_serial)
                for i in range(0, len(toks), 2):
                    idx[int(toks[i])] = int(toks[i + 1])
            elif line.startswith("REMARK SMILES"):
                smiles = line.split("REMARK SMILES", 1)[1].strip()
    return smiles, idx                                # idx[smiles_atom_position] -> ligand atom serial

for f in glob.glob("*_cmpx.pdb"):
    smi, idx = read_complex(f)
    # ligand atoms are the `ATOM ... UNL` records following `MODEL 1`

Caveats

• —C6-ceramide is a synthetic short-chain surrogate; physiological ceramides carry C16–C24 acyl chains. The enclosed acyl subsite used as the growth vector is defined by a 6-carbon stub spanning only 5.8 Å, so its true extent is not established by this structure.
• —All nine ORMDL3 pocket contact positions are identical in ORMDL1 and ORMDL2 — paralog-selective chemistry is not achievable at this site. Design for pan-ORMDL and manage exposure by inhaled, lung-restricted delivery.
• —The direction of pharmacology is unproven, and no HTS-ready SPT-activity assay currently exists.
• —The causal gene at 17q21 is contested; GSDMB, IKZF3 and ZPBP2 share the haplotype block.

Provenance & intended use

• —These are computationally generated designs and docked poses — not experimentally validated binders. No claim of activity or selectivity is made.
• —Target proposal, site definition and pharmacophore modelling by Claude Code. Molecule generation by the Technetium `TC-43.ai` engine.
• —Intended for machine-learning, cheminformatics, generative-model benchmarking, and docking-pose research on a well-defined, structurally characterised target.

Citation

Generated by Technetium Therapeutics. Poses scored with AutoDock Vina. Receptor derived from RCSB PDB 7YIU.