CD38 is the dominant NAD-consuming ectoenzyme in aged tissues; its activity rises ~2–3 fold with age, suppressing the NAD+ pool that fuels sirtuins and PARP. Structural groundwork from PDB:2I65 (1.9 Å X-ray, human CD38 ecto-domain) and the catalytic hotspot geometry profile (C119, D155, D156, D179, D219, Q226; 6 hotspots confirmed present, 8 neighbor residues at ≤4.5 Å) establishes the conditioning target for RFdiffusion backbone generation. Planned pipeline: (1) RFdiffusion binder generation conditioned on 2I65 chain A hotspot centroid coordinates, 50–80 residue miniprotein scaffolds, partial diffusion with hotspot conditioning; (2) ProteinMPNN sequence design on top-20 backbones by predicted RMSD; (3) ESM-2 log-likelihood filter (ppl < 10) to remove low-confidence sequences; (4) AlphaFold2-Multimer forward-fold validation of CD38:binder complex (ipTM > 0.7, pLDDT > 80 on binder, PAE < 10 Å at interface) — AF2 used as oracle independent of RFdiff design model; (5) Rosetta FastRelax + InterfaceAnalyzer ddG, dSASA, shape complementarity; (6) Foldseek structure search against PDB to assess novelty and flag any unintended scaffold matches to known immunogenic folds; (7) selectivity check: AF2-Multimer against CD157 (BST1, structurally related) to confirm binder does not cross-react. Prior art baseline: small-molecule inhibitor 78c (IC50 ~4 nM on mouse CD38; human CD38 Ki ~7 nM from Camacho-Pereira et al. 2016 Cell). Success criterion: AF2-validated design with ipTM > 0.75, predicted ddG < −10 kcal/mol, Foldseek TM-score < 0.7 to any known antibody CDR scaffold.
Details
- disease
- aging / nad+ decline / metabolic aging
- target_ref
- UniProt:P28907 / PDB:2I65
- primary_endpoint
- At least 1 of 10 BindCraft designs achieves ipTM ≥ 0.70 in the BindCraft complex prediction
- identification_strategy
- in_silico_KO
Raw fields (3)
- assay_spec
Parse PDB 2I65 chain A catalytic pocket (C119, D155, D156, D179, D219, Q226), extract hotspot geometry JSON, then run BindCraft de novo binder design conditioned on those 6 hotspot residues, followed by ESM-2 perplexity filter (pseudo-perplexity < 8) on top-10 sequences; hotspot_cd38_2i65.json — structured hotspot input file for BindCraft: list of (chain, resnum, resname) tuples for C119/D155/D156/D179/D219/Q226 with centroid coordinates; bindcraft_cd38_top10.fasta — top-10 BindCraft-designed binder sequences ranked by predicted ipTM; all sequences 60–120 aa, no free cysteines; esm2_perplexity_filter.csv — ESM-2 pseudo-perplexity scores for all 10 sequences; sequences with perplexity > 8 flagged for removal; {'kind': 'claim', 'text_template': '≥1 BindCraft binder candidate achieves BindCraft ipTM ≥ 0.7 against CD38 2I65 chain A hotspot and ESM-2 perplexity < 8'}; Lead candidate FASTA + predicted complex PDB + ipTM/pAE table — registered as protein_design artifact linked to research_plan:5e4168c3- hypothesis
A de novo protein binder designed to occlude the CD38 catalytic cleft (hotspot residues C119, D155, D156, D179, D219, Q226 from PDB:2I65) will competitively inhibit NAD+ hydrolysis with selectivity over structurally related ectoenzymes, thereby elevating NAD+ in aged tissues and attenuating age-associated metabolic decline.
- kill_criteria
Abort or revise if: BindCraft compute availability — if GPU cluster is unavailable, fallback is to use existing protein_design artifacts (54744156, df6b98ee, e658dbb9, c340aeb5) and run ESM-2 perplexity filter only; 2I65 has two chains (A/B homodimer) — must isolate chain A only to avoid designing binders that occlude the dimer interface rather than the catalytic pocket; CD38 active site is occluded by substrate analog in 2I65 — verify pocket accessibility (apo vs. holo conformation) before treating hotspot as fully druggable surface