Tc-43/LHCGR_Agonist_Designs_7FIH

LHCGR Allosteric Agonist Designs — 7FIH (GA-II)

993 small molecules generated de novo by the Technetium `TC-43.ai` engine (GA-II), conditioned on the transmembrane allosteric pocket of the luteinizing hormone / choriogonadotropin receptor (LHCGR), using PDB 7FIH as the receptor. 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. All 993 are profiled for protein–ligand interactions.

Target proposal and structural modelling by Claude Code. Molecule generation by the Technetium `TC-43.ai` engine (GA-II).

Companion set: FSHR_Agonist_Designs_8I2G — the two receptors are each other's counter-screen.

Where the target came from

On 9 September 2026, EMD Serono announced the acquisition of PostEra's fertility programmes — FSHR agonists for ovarian stimulation and LHCGR agonists for male infertility and ovulation trigger (press release). Both programmes are preclinical; no financial terms were disclosed. From the release:

"Both FSHR and LHCGR are targets which have been validated by biologics, but over the past few decades there has been little progress in drugging them with orally bioavailable small molecules."

That announcement is the only input this programme started from. Claude Code took the target names from the news, selected the structures, built the pocket and selectivity analysis, and specified the docking setup the same day. The generated, docked and interaction-profiled design sets for both receptors were complete four days later.

The timeline is the point: a public disclosure to a structure-resolved, selectivity-analysed design set in days, driven from natural language. What that demonstrates is the speed and reach of the design loop — not that these molecules work. Nothing here has been synthesised or assayed.

Why this target

Oral small-molecule LHCGR agonism is clinically proven in humans. Org 43553 (ligand 55Z in 7FIH) induced ovulation in a randomised placebo-controlled single-rising-dose study in healthy female volunteers: minimum effective dose 300 mg, ovulation rate 83%, well tolerated. The programme is dormant — no registered studies — so the biology is de-risked and the chemistry is open.

Paired with an FSHR agonist, this is the trigger half of an all-oral IVF protocol; alone it is a candidate for male infertility.

The reference tells you what not to optimise. Org 43553 is 3.7 nM at LHCGR and still needed 300 mg. Affinity was never the limiting property. What is limiting:

So this set was assessed on FSHR/TSHR selectivity and oral-property quality, not on docking score alone.

Structure and pocket

7FIH — LHCGR(S277I) + hCG + Gs + Org 43553, cryo-EM, 3.2 Å. Receptor chain R, residues 52–644. Docking box centre (137.22, 153.16, 155.38), size 14 × 18 × 22 Å.

Of the 23 pocket residues, 16 are identical in FSHR. The seven that differ:

What the set does

It found the selectivity handle the reference misses. Org 43553 touches Phe515 only as a hydrophobic wall contact. 493 of 993 designs (49.6%) make a genuine aromatic interaction with Phe515 — 473 π-stacks and 38 cation-π. Phe515 is the single position where LHCGR differs from both FSHR and TSHR, and an aromatic stack there has no counterpart against a leucine.

It binds deeper than the reference. Mean ligand-centroid displacement from the 55Z pose is 2.8 Å, directed into the TM core. The designs pick up Leu532, Leu607, Leu608 and Tyr527, which Org 43553 never reaches.

It gives up the Lys595 arm. Org 43553's morpholine-acetamide reaches Lys595 — its most LHCGR-specific feature. Only 29 designs (2.9%) engage Lys595 at all, concentrated in chemotype cluster 10. That subpocket is unexploited and is the clearest place for the next generation to grow.

Interaction profile across all 993 (Tc_PLIP, default profile):

The Ile585 backbone H-bond in 53% of designs sits on the TM6 face that moves on activation — the surface reproducing Org 43553's contact pattern most closely.

Reference for comparison — 55Z re-profiled in the same pocket with the same settings: 2 H-bonds (Ser604 2.78 Å, Lys595 3.41 Å), 5 hydrophobic contacts (Ile531, Met517, Phe515, Ala589, Val447), no π-stacking.

Selectivity tiers

selectivity_handles counts engagement of the four positions that differ from FSHR or TSHR: aromatic Phe515, Ser604 H-bond, Ala593 contact, Val519 contact.

32 designs combine an aromatic Phe515 interaction with a Ser604 H-bond — the two-point pharmacophore that is differentiated against FSHR at both positions.

Properties

Median MW 410, cLogP 2.9, TPSA 88, HBD 3, Fsp³ 0.45, QED 0.66. 99.8% pass Ro5. Against Org 43553 (MW 515, QED 0.34) the set is ~100 Da lighter with roughly double the drug-likeness score, and ligand efficiency runs 0.31–0.51 (median 0.41).

83 designs fall inside the strict Org 43553 envelope (MW ≤ 520 · cLogP 2.5–4 · TPSA 90–130 · HBD 2–3 · Fsp³ ≥ 0.4); the binding constraint is TPSA, where the set skews low (median 88 vs a 90–130 window) because the molecules are smaller.

Novelty

733 unique Murcko scaffolds across 993 designs; 209 Butina clusters at Tanimoto 0.4, of which 114 are singletons. Maximum ECFP4 similarity to Org 43553 is 0.19 — no design is a thieno[2,3-d]pyrimidine analogue. The recurring motif is a 3,4-dihydro-2H-1,4-benzoxazine biaryl, unrelated to the reference chemotype.

Files

Key columns

• —priority_rank — composite of Phe515 aromatic contact, Ser604 H-bond, Ala593/Val519 contacts, docking score, QED, ligand efficiency and property-envelope fit
• —selectivity_handles — 0–4, FSHR/TSHR-discriminating positions engaged
• —F515_aromatic, hb_S604, hb_Y612, hb_I585, ph_A593, ph_V519, any_K595 — binary
• —centroid_offset_A — ligand centroid displacement from the Org 43553 pose
• —contact_residues — all residues within 4.5 Å; plip_residues — residues with a classified interaction
• —chemotype_cluster — Butina cluster at Tanimoto 0.4

Load

import pandas as pd
df = pd.read_csv("hf://datasets/Tc-43/LHCGR_Agonist_Designs_7FIH/designs.csv")

# the two-point FSHR-differentiated pharmacophore
df.query("F515_aromatic == 1 and hb_S604 == 1").head(32)

Caveats

• —Docking predicts binding, not agonism. No score distinguishes an agonist from an antagonist. The partial mitigation here is that Org 43553 is a functional agonist validated in humans, so reproducing its contact pattern — the TM6/TM7 face at Ile585/Phe588/Ala589/Ala592/Ala593 — is a more defensible activation proxy than anything available on the FSHR side.
• —3.2 Å is coarse for rotamers. Weight backbone geometry and pocket shape over fine side-chain detail; refine the site before committing to close SAR.
• —The construct is S277I, an activating mutation 14.1 Å from the ligand — outside the pocket, but the receptor is biased toward the active state by design.
• —Rigid receptor, single pose per design. No induced fit, no ensemble.
• —Selectivity is inferred from sequence, not computed. FSHR (8I2G) and TSHR (7T9M) counter-docking has not been run on this set.
• —Nothing here has been synthesised or tested. The acquired programmes referenced above are preclinical; safety and efficacy in humans have not been established.
• —Some SMILES carry explicit protonation states from the generator.

Citation

@dataset{technetium_lhcgr_7fih_gaii,
  title  = {LHCGR Allosteric Agonist Designs (7FIH, GA-II)},
  author = {TechnetiumTx},
  year   = {2026},
  url    = {https://huggingface.co/datasets/Tc-43/LHCGR_Agonist_Designs_7FIH}
}

License: CC-BY-4.0.