Tc-43/nras-cypa-macrocyclic-glues-GA-II

NRAS–Cyclophilin A Macrocyclic Glue Designs (GA-II)

Why this target matters. NRAS-mutant melanoma has no approved targeted therapy and poor outcomes once immunotherapy fails; RAS(ON) tri-complex glues are among the very few mechanisms that engage NRAS at all.

180 small molecules generated de novo by the Technetium TC-43.ai engine (GA-II), conditioned on the NRAS·Cyclophilin A protein–protein interface, with macrocyclic ring closure imposed during generation. 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 contacts both proteins simultaneously — the defining geometry of a RAS(ON) molecular glue — and each is delivered as a full three-body complex structure. No screening library covers this beyond-Rule-of-5 macrocyclic space.

The receptor is the ternary NRAS(GMPPNP)·CypA assembly from PDB 9BG0, the structural template for tri-complex RAS(ON) inhibitors such as daraxonrasib (RMC-6236).

What's in it

Designs	180 macrocycles (179 unique structures)
Ring size	20–27 atoms (median 21)
Docking score	−13.24 to −10.00 kcal/mol (median −10.49)
Interface span	180/180 contact both NRAS and CypA
Scaffold diversity	179 unique Murcko scaffolds; 155 Butina clusters at Tc 0.65
Chemical space	MW 510–729, cLogP 0.0–6.0, TPSA 76–188, Fsp³ 0.2–0.6

All 180 sit in beyond-Rule-of-5 oral space; none pass classic Ro5 — expected and appropriate for macrocyclic glues.

Interface engagement

Contacts were computed at a 4.5 Å heavy-atom cutoff. Numbers in parentheses are how many of the 180 poses touch that residue.

Cyclophilin A — the designs occupy the PPIase active site, the same pocket used by cyclosporin A and sanglifehrin:

Phe60 (179), Arg148 (164), Ile57 (162), Trp121 (156), Met61 (148), Leu122 (136), Arg55 (124), Gln63 (104), His126 (101), Phe113 (79)

NRAS — engagement centres on Switch II with a consistent Switch I foothold:

Tyr64 (180), Ile36 (178), Ala59 (169), Met67 (168), Gln61 (145), Thr35 (142), Glu37 (125), Pro34 (76), Tyr71 (68), Gln70 (36)

Only 5 poses reach the GMPPNP nucleotide or Mg²⁺, so the series binds the induced interface groove rather than competing with the nucleotide.

A shared N-methylindole–thiazole–propanamide arm recurs across the series as the CypA anchor, with diversification concentrated on the NRAS-facing half of the macrocycle.

Files

data/designs.parquet             # 180 rows × 23 columns (main table)
designs.csv                      # same, flat columns, no contact lists
structures/<design_id>_complex.pdb   # 180 three-body complexes
receptor_9BG0_NRAS_CypA.pdb      # shared receptor (rigid across all poses)

Columns

Column	Description
rank, design_id	Rank by docking score; unique design identifier
smiles	Design structure
vina_kcal_mol	AutoDock Vina score
mw, clogp, tpsa, hbd, hba, rotatable_bonds, heavy_atoms, fsp3	RDKit physicochemical properties
n_rings, max_ring_size, n_aromatic_rings	Ring topology; max_ring_size is the macrocycle
murcko_scaffold	Murcko scaffold SMILES
n_contacts_nras, n_contacts_cypa	Residue counts within 4.5 Å
contacts_nras, contacts_cypa, contacts_nucleotide	Residue identity lists
spans_interface	True when the design contacts both proteins
structure_file	Path to the complex PDB

Each complex PDB carries chain A (NRAS + GNP + Mg²⁺), chain D (CypA), and the ligand as LIG, with REMARK SMILES and CONECT records.

Usage

from datasets import load_dataset
ds = load_dataset("Tc-43/nras-cypa-macrocyclic-glues-GA-II", split="train")
ds = ds.sort("vina_kcal_mol")
print(ds[0]["smiles"], ds[0]["vina_kcal_mol"])

# pull the top-ranked complex structure
from huggingface_hub import hf_hub_download
p = hf_hub_download("Tc-43/nras-cypa-macrocyclic-glues-GA-II", ds[0]["structure_file"], repo_type="dataset")

Caveats

These are computational designs, not experimental structures. Docking scores rank geometric and steric complementarity; they are not measured affinities, and no design here has been synthesised or assayed. The receptor is held rigid, so induced fit at the interface is not modelled, and the docking score does not account for the entropic cost of macrocyclic preorganisation or the cooperativity that defines true glue behaviour. Treat the interface contact patterns as the primary signal and the scores as a coarse filter.

Citation

@misc{nras_cypa_ga2_macrocycles,
  title  = {NRAS--Cyclophilin A Macrocyclic Glue Designs (GA-II)},
  author = {TechnetiumTx},
  year   = {2026},
  url    = {https://huggingface.co/datasets/Tc-43/nras-cypa-macrocyclic-glues-GA-II}
}