Quantum annealing-based subgraph isomorphism algorithms can identify structural motifs in protein-ligand docking data that correlate with transcriptomic biomarkers of Multiple Sclerosis severity.

This discovery has a Claude-generated validation package with a full experimental design.

Precise Hypothesis

Spine & Adversarial Read

“This hypothesis tests whether structural binding-pocket motifs identified by quantum-annealing subgraph isomorphism in CTSS/DNMT1/BRD3 docking data statistically correlate with the transcriptomic severity metrics already established for these targets in multiple sclerosis.”

• highThe transcriptomic 'severity biomarker' being correlated against is a single fixed composite score per target (0.578 for CTSS, etc.), not a per-ligand or per-patient variable — correlating a distribution of per-compound motif scores against one scalar is statistically meaningless as framed; the analysis needs reframing as per-patient severity vs. drug-response, which requires data not currently specified as available.
  Not resolved in this EVP. Requires either (a) patient-stratified drug response data linking CTSS inhibitor exposure to clinical/transcriptomic severity, or (b) reframing the correlation as motif-score vs. ChEMBL potency (a valid but biologically weaker claim that doesn't actually test MS-relevance). This is flagged explicitly as a required design fix before experimental funding, not merely a risk.
• highWhy quantum annealing specifically, rather than classical exact/approximate subgraph isomorphism (VF2, RI, gSpan) or GNN-based motif learning, which are cheaper, faster, and already standard in cheminformatics? The methodology does not justify a quantum-specific computational necessity for graphs of this size (pocket graphs ≤200 nodes are classically tractable).
  Partially resolved: protocol includes mandatory classical baseline comparison (step 6, success/failure criteria include explicit quantum-vs-classical benchmarks) and an abort checkpoint (Day 55) if no advantage is found. However, the EVP acknowledges rather than pre-establishes quantum necessity — this is an open empirical question the protocol is designed to answer, not a justified premise.
• mediumZNF740 has no solved 3D structure, so using BRD3 as a docking proxy conflates two different proteins' pharmacology; any motif-transcriptome correlation found for BRD3 cannot be validly attributed to ZNF740's disease mechanism.
  Acknowledged directly in boundary conditions — the EVP restricts ZNF740 claims to 'BRD3 as druggable proxy' and does not claim direct ZNF740 structural validation. This narrows the ZNF740 arm's interpretability but does not invalidate the CTSS/DNMT1 arms, which have direct structures.

Closest Existing Work

• Transcriptomic and network medicine analysis of chronic active rim lesions in multiple sclerosis identifies CTSS, ZNF740, and DNMT1 as candidate therapeutic targets — Goodman J. (2026)

  Established: Phase 3 multi-database validation combined transcriptomic expression, GEO replication (GSE138614), STRING proximity, and DrugBank actionability to identify CTSS (composite 0.578), ZNF740 (0.648), and DNMT1 (0.912) as MS smoldering-lesion therapeutic targets.

  Novelty after this: Self-validation: this EVP proposes to interrogate the same three targets via QA-SGI. The composite scores used as correlation reference are derived from this work, making it both the enabling foundation and the primary comparator.

Experimental Protocol

ROI Projection

Implementation Sketch

for target in [CTSS, DNMT1, BRD3]:
    ligands = fetch_chembl_ligands(target, pChEMBL>=5, n>=50)
    poses = dock_or_retrieve(ligands, target_structure=PDB[target])
    graphs = [build_interaction_graph(pose, cutoff=6A) for pose in poses]

    qubo = formulate_subgraph_isomorphism_QUBO(graphs, motif_size_range=(4,12))
    qa_result = dwave_hybrid_solver.sample(qubo, num_reads=1000)
    classical_result = vf2_baseline(graphs, motif_size_range=(4,12))

    motifs_qa = extract_top_k_motifs(qa_result, k=10)
    motifs_classical = extract_top_k_motifs(classical_result, k=10)

    validate_against_cocrystal(motifs_qa, known_pharmacophores[target])  # F1 score

    activation_scores = score_ligands_by_motif_overlap(ligands, motifs_qa)
    transcriptomic_metric = load_composite_score(target)  # from preprint

    rho, fdr = spearman_corr_with_fdr(activation_scores, transcriptomic_metric)
    partial_rho = partial_corr(activation_scores, transcriptomic_metric,
                                controls=[MW, LogP, docking_score])

    compare_qa_vs_classical(qa_result, classical_result, metric=['F1','walltime','chain_break_rate'])