Spoke 03 // AI Drug Discovery, Perturbation Omics & Structural Design
Which Orphan Indications Give AI-Bio Teams Both a Clean Structural Selectivity Target and Open Human Single-Cell Data?
JAK2 V617F+ myelofibrosis offers AI-bio teams two rare computational advantages with $0 upfront wet-lab capex: (1) a well-posed structural allostery problem—sparing essential wild-type JAK2 by targeting the JH2 pseudokinase V617F switch rather than the conserved JH1 catalytic pocket—and (2) open human bone marrow single-cell datasets (GSE144568, GSE122198) resolving mutant vs. wild-type stem cells within the same patient.
Why Do Most "AI Drug Repurposing" Projects Stall—and Why Does JAK2+ MF Avoid Those Traps?
In-silico drug repurposing and generative design projects typically fail translation when human primary tissue is inaccessible, bulk RNA-seq confounds cell-type proportions, or ex-vivo validation requires multi-year animal models. Evaluate JAK2 V617F+ myelofibrosis across computational readiness criteria:
| Translational Requirement | Typical Rare Solid/Neuro Disease | JAK2 V617F+ Myelofibrosis Reality |
|---|---|---|
| Single-Cell Mutant vs. WT Ground Truth | Rare; bulk biopsies mix stroma, immune infiltrate, and heterogeneous subclones. | Solved Openly GSE122198 (TARGET-seq) pairs single-cell JAK2 V617F genotyping with full transcriptomes inside individual marrows. |
| Structural Selectivity Hypothesis | Disordered proteins or unclear loss-of-function rescue. | Atomic Resolution Gain-of-function V617F aromatic stacking in JH2 pseudokinase disrupts JH1 autoinhibition & drives cytokine-independent dimerization. |
| Fast CRO Wet-Lab Validation Loop | Slow organoids or unfaithful murine models. | 2–3 Week Readout Ba/F3-EPOR-JAK2(V617F vs. WT) isogenic IC50 selectivity screens + primary human MF CD34+ colony-forming assays via standard hematology CROs. |
| Clinical Combination Backbone | Unclear standard of care. | Clear Add-On Path Repurposed non-myelosuppressive hits can layer directly onto genericizing ruxolitinib or ACVR1-sparing momelotinib. |
What Can an ML Engineer or Computational Biologist Build Starting Tonight?
Challenge A: Wild-Type-Sparing JH2 Allosteric Binders & Mutant-Selective Degraders
Every approved MF JAK inhibitor (ruxolitinib, fedratinib, pacritinib, momelotinib) is an orthosteric Type I inhibitor competing with ATP in the catalytic JH1 domain, which is sequence-identical between mutant and wild-type JAK2. Consequently, systemic exposure sufficient to wipe out mutant stem cells causes severe anemia (blocking EPO-JAK2) and thrombocytopenia (blocking TPO-JAK2). Furthermore, Type I inhibition causes paradoxical JAK2 activation-loop phosphorylation accumulation ("persistence").
- Computational Objective: Model full-length homodimeric receptor-JAK2 complexes (EpoR/TpoR + FERM-SH2-JH2-JH1) across wild-type vs.
V617Fconformational ensembles. Design small molecules stabilizing the autoinhibitory JH2–JH1 interface exclusively in the presence of Phe617, or design VHL/CRBN molecular glues/PROTACs exploiting JH2 V617F ternary cooperativity.
Challenge B: Single-Cell Perturbation Reversal of Megakaryocyte-Driven Bone Marrow Fibrosis
In GSE144568 (Psaila et al., Molecular Cell 2020; companion code at SingCellaR-myelofibrosis), single-cell profiling of lineage-negative CD34+ HSPCs demonstrated that mutant stem cells bias early toward aberrant megakaryocyte progenitors secreting profibrotic mediators (TGFB1, PF4, PDGFA, LTBP1) that remodel bone marrow mesenchymal stromal cells into collagen-depositing myofibroblasts.
- Computational Objective: Use single-cell foundation models and LINCS/Tahoe-scale chemical perturbation signatures to identify FDA-approved, non-myelosuppressive oral compounds that selectively collapse the aberrant megakaryocyte-fibrosis transcriptional program without suppressing normal erythroid progenitor differentiation.
Challenge C: Synthetic Lethality Against High-Molecular-Risk (HMR) Subclonal Escape
Ruxolitinib discontinuation and leukemic transformation to secondary AML are heavily accelerated by co-mutations in chromatin and splicing regulators—specifically ASXL1, EZH2, SRSF2, U2AF1, and IDH1/2. Using single-cell multi-mutational transcriptomes (GSE122198), computational teams can map differential isoform usage (from SRSF2/U2AF1 neomorphic splicing) to discover neoantigens for bispecific T-cell engagers or synthetic-lethal metabolic bottlenecks unique to double-mutant (JAK2 V617F + ASXL1/SRSF2) clones.