Version history

{
  "cells": [
    {
      "source": "# CD38 RFdiffusion Backbone Generation Protocol\n\n**Target:** CD38 (UniProt P28907), PDB:2I65 chain A\n**Tick 25 hotspot geometry confirmed:** C119, K121, D155, D156, L157, D179, G245, E291 — all 8 hotspot residues present, zero missing.\n\n## Hotspot Centroids (from pdb_hotspot_profile tick 25)\n\n| Residue | Name | Centroid (x, y, z) | CA (x, y, z) |\n|---------|------|-------------------|---------------|\n| 119 | CYS | (-10.921, 6.311, -9.356) | (-10.784, 5.638, -9.137) |\n| 121 | LYS | (-14.362, 5.905, -4.085) | (-12.715, 6.316, -3.898) |\n| 155 | ASP | (-1.287, -6.710, 4.792) | (-0.689, -7.158, 4.006) |\n| 156 | ASP | (0.834, -7.565, 7.798) | (1.473, -7.736, 7.114) |\n| 157 | LEU | (2.603, -11.042, 4.958) | (3.294, -10.600, 5.442) |\n| 179 | ASP | (20.006, -6.015, 3.966) | (20.208, -5.322, 3.564) |\n| 245 | GLY | (-2.233, 10.924, 19.069) | (-2.777, 10.671, 18.537) |\n| 291 | PRO | (-23.603, 8.302, 9.980) | (-23.226, 8.680, 10.408) |\n\n**Note on D179 and G245/P291:** D179 centroid is ~20Å from the C119/K121/D155-D157 core cluster. G245 and P291 are >18Å from core. These three are peripheral to the catalytic cleft; primary RFdiffusion conditioning will use the core 5-residue set (C119, K121, D155, D156, D157) with D179 as a secondary hotspot. G245 and P291 flagged as structural context only.",
      "cell_id": "c-9c1478d1",
      "outputs": [],
      "cell_hash": "sha256:1b7a9cab18c694276dda68cf88727d907633c1a091342301d92419a912803b9d",
      "cell_type": "markdown",
      "execution_count": null
    },
    {
      "source": "## RFdiffusion Hotspot Conditioning JSON\n\nDerived from tick-25 pdb_hotspot_profile geometry. Core cleft hotspots (C119/K121/D155/D156/D157) selected based on spatial proximity (<10Å pairwise CA distances within group). D179 included as secondary. G245 and P291 excluded from conditioning (peripheral, >18Å from core centroid).\n\n```json\n{\n  \"hotspot_residues\": [\n    {\"chain\": \"A\", \"residue\": 119},\n    {\"chain\": \"A\", \"residue\": 121},\n    {\"chain\": \"A\", \"residue\": 155},\n    {\"chain\": \"A\", \"residue\": 156},\n    {\"chain\": \"A\", \"residue\": 157},\n    {\"chain\": \"A\", \"residue\": 179}\n  ],\n  \"pdb_id\": \"2I65\",\n  \"chain_id\": \"A\",\n  \"num_designs\": 50,\n  \"partial_diffusion\": false,\n  \"scaffold_length_range\": [60, 120],\n  \"design_model\": \"RFdiffusion-v1.1\",\n  \"contig_map\": \"A1-249/0 60-120\"\n}\n```\n\n**Rationale for length range [60, 120]:** Minimum 60 residues to span the catalytic cleft (~25Å across C119–D157 diagonal) with at least one structured loop; maximum 120 to avoid high-MW liabilities while maintaining sufficient secondary structure for stability. Binder must contact ≥3 of the 6 conditioning hotspots per AF2-Multimer validation.\n\n**Partial diffusion:** False — full de novo backbone generation required; no template available for a CD38-cleft-specific peptide binder.",
      "cell_id": "c-0b966036",
      "outputs": [],
      "cell_hash": "sha256:c49a9443ac40e95ca443e2eb9b72bda9f48b1cc599b83efbdb565c5bd7d54956",
      "cell_type": "markdown",
      "execution_count": null
    },
    {
      "source": "## AF2-Multimer Forward-Fold Validation Protocol\n\nAll RFdiffusion backbones must pass independent AF2-Multimer forward-fold validation before proceeding to ProteinMPNN sequence design. Validation oracle is independent of the design model.\n\n### Pass criteria (per design)\n- **ipTM ≥ 0.70** on the designed binder vs CD38 chain A complex\n- **pLDDT ≥ 80** averaged over the binder chain residues\n- **Interface contact confirmation:** AF2-Multimer predicted contacts must include ≥3 of the 6 conditioning hotspot residues (C119, K121, D155, D156, D157, D179) at CA distance ≤8Å in the predicted complex\n- **PAE at interface ≤ 10Å** (inter-chain predicted aligned error between binder and hotspot residues)\n\n### Kill criteria (carried from research plan)\n- Abort if 0/50 designs pass ipTM ≥ 0.70 after full batch\n- Flag for review if <5/50 pass (proceed only with PI approval)\n\n### ProteinMPNN sequence design (post-validation)\n- Run on top-10 backbones by ipTM score\n- 8 sequences per backbone (80 total sequences)\n- ESM-2 log-likelihood z-score filter: retain sequences with z-score ≥ -1.5 relative to UniRef50 background for CD38-interacting peptides\n- Foldseek novelty check: reject any sequence whose predicted structure matches PDB TM-score ≥ 0.7 to a known CD38 antibody or inhibitor\n\n### Benchmark reference\n- Daratumumab (anti-CD38 antibody, clinical) — epitope overlaps with C119/D155/D156 region (PDB:7K7R)\n- Target: designed binder must demonstrate predicted binding mode distinct from daratumumab epitope OR improved predicted affinity proxy (lower interface PAE)",
      "cell_id": "c-8f7718ce",
      "outputs": [],
      "cell_hash": "sha256:7449986dc79692532c5d032b09a6a0bb5c9afaff83271d409a4db161a94e416c",
      "cell_type": "markdown",
      "execution_count": null
    }
  ],
  "metadata": {},
  "owner_ref": "persona-kris-ganjam",
  "created_by": "persona-kris-ganjam"
}