Where the engine points next.

AKT1 is the one surviving synthetic-lethal co-dependency of PIK3CA-mutant lines (DepMap 24Q4; PIK3CB/MTOR/PDPK1/GLS/RNMT refuted). The AKT1 receptor is redock-verified (capivasertib 1.83Å). Next novel-modality bets: AKT1 degrader, E17K allele-selective.

The mutant-p110α degrader inavolisib ENGAGES the redock-verified canine PIK3CA ATP pocket (docking supports; engagement != degradation). Next: orthogonal cofolding, then ternary/E3 modeling. Grounded on a no-substitution canine target.

KDR is our most heavily validated axis: multiple small molecules co-fold cleanly into a stable, engageable KDR structure (lenvatinib cofolding iptm 0.98 / P(binder)=0.89, tivozanib P(binder)=0.85, axitinib P(binder)=0.86), and lenvatinib docks at -10.1 kcal/mol. That establishes a real, foldable pocket a constrained peptide can be cofolded against. A stapled peptide adds proteolytic stability and a larger interaction surface than the ATP-competitive TKIs already crowding this target, which is the frontier differentiator. Fits the cofolding lane directly.

The copanlisib warhead engages the canine PIK3CA pocket (lower pose confidence, honest). To advance: a CRBN/VHL ternary-complex lane + the full-PROTAC structure.

PIK3CA folds and engages ligands robustly in our runs: alpelisib docks -9.4 and co-folds at iptm 0.99 / P(binder)=0.91, inavolisib co-folds P(binder)=0.97, taselisib affinity_pred -2.47. That confirms a well-behaved structure a designed peptide can be cofolded into. A stapled peptide can reach the helical-domain surface (site of the E545K hotspot) that small molecules poorly address, making it a genuine frontier play against a validated driver. Cofolding lane.

Docking scores pocket engagement on a wild-type receptor only — not degradation and not mutant-selectivity. These are the honest next capabilities, not yet built.

We already have a covalent-modality foothold on this exact target: CNX-1351 (covalent PI3Kalpha, Cys862) docked -9.2/-9.6 kcal/mol with a supporting signal, and reversible binders alpelisib (-9.4) and inavolisib (-9.5) confirm the ATP pocket geometry the warhead must reach. A covalent binder converts that transient engagement into durable, sub-stoichiometric inhibition and can be tuned for mutant selectivity. Directly dockable, cheapest strong bet.

Two mutant-selective allosteric binders establish a druggable non-orthosteric surface on PIK3CA: STX-478 docked -7.3/-9.2 and RLY-2608 docked -9.5/-10.4, both supporting. That allosteric pocket is exactly where a molecular glue can nucleate a ternary complex to degrade only the oncogenic mutant, sparing wild-type p110alpha. Alpelisib's clean co-fold (P(binder)=0.91) anchors the receptor model for ternary cofolding. Testable via cofolding of the glue-bridged complex; higher-risk so priced/confidence accordingly.

ZSTK474 was explicitly logged as a PI3K-PROTAC warhead and docked into PIK3CA at -8.4/-8.8 kcal/mol (CNN pose 0.95), giving us a real, ledger-proven anchor for the target-binding half of a degrader. Alpelisib (-9.4) corroborates the pocket. Peptidic E3-recruiting linker turns engagement into catalytic degradation. The warhead-binding arm is directly dockable now.

Alpelisib is our strongest, most reproducible PIK3CA binder (docking -9.4, co-fold iptm 0.99 / P(binder)=0.91, still-standing across replicate candidates), making it an ideal, validated warhead to convert into a degrader. Inavolisib (-9.5) confirms the pocket tolerates diverse chemotypes for exit-vector placement. Degradation removes scaffold/kinase-independent p110alpha functions that occupancy inhibitors miss. Warhead conjugate is dockable now.

The KDR ATP pocket is validated by strong, supporting docks (cabozantinib -9.4/-11.2, lenvatinib -10.1, toceranib -7.2 with clean co-fold). Any of these gives a proven warhead for the target-binding arm of a peptide-PROTAC; the peptidic E3 recruiter converts inhibition into degradation of the receptor, a differentiated mechanism versus the many occupancy TKIs already tested here. Warhead arm is dockable now.

PRN1371 (a covalent kinase tool) only reached a neutral -6.6/-6.9 dock against KDR and is logged needs-more, so the covalent arm on this target is under-optimized rather than refuted. Meanwhile reversible binders axitinib (-11.4) and regorafenib (-11.5/-12.3) prove the pocket can be engaged very tightly, defining the affinity headroom a better covalent scaffold should recover before making the bond irreversible. Directly dockable, cheap re-attempt with a clear falsifiable bar.

The target itself is validated as the disease-relevant driver (alpelisib co-folds P(binder)=0.91; inavolisib P(binder)=0.97), and the mutant stratum is real in our cohort (PIK3CA-mutant n=5 vs WT n=9 in the omics crux). A base edit does not bind a pocket and cannot be docked or co-folded; it must be assessed by editable-window/PAM feasibility, bystander analysis, and delivery literature. Explicitly a non-structural evaluation, so no compute lane and confidence kept low.

KDR is a validated, engageable cell-surface receptor here (multiple supporting co-folds: toceranib P(binder)=0.73, cabozantinib 0.80, lenvatinib 0.89), and its surface localization is what makes it a CAR antigen. But a CAR recognizes an extracellular epitope through an scFv; there is no small-molecule pocket to dock or co-fold. Evaluation must be non-structural: antigen surface density, on-target/off-tumor vascular toxicity, and canine feasibility. No compute lane.

PIK3CA is a validated driver (alpelisib docks -9.4) and a recurrent hotspot mutation is the kind of shared neoantigen mRNA vaccines exploit. Critically, our own omics crux found the PIK3CA-mutant stratum trends more immunosuppressed (neutral, 0/7 significant), which is a real caveat this modality must confront, not a green light. A vaccine has no dockable structure; it requires MHC-binding prediction, immunopeptidome and TME evaluation. Non-structural, low confidence, no compute lane.

KDR is validated as an engageable surface receptor (lenvatinib co-fold P(binder)=0.89, tivozanib 0.85, toceranib 0.73), and receptor-mediated internalization is the mechanism an ADC needs. But the antibody arm and payload act through binding/internalization/linker chemistry, none of which is a small-molecule pocket to dock or co-fold. Evaluation is non-structural: internalization rate, payload/linker choice, endothelial toxicity. No compute lane.

The PIK3CA driver is validated by clean, still-standing engagement (alpelisib co-fold P(binder)=0.91, inavolisib 0.97), justifying a direct genetic correction. Prime editing offers precise reversion without the bystander limits of base editing, but like all gene edits it has no bindable structure to dock or co-fold; assessment is pegRNA/PBS design, editing efficiency and delivery. Non-structural, no compute lane, confidence low.

The PIK3CA-mutant subset is a real, defined stratum in our cohort (n=5 vs WT n=9 in the omics crux) and dactolisib confirms the pathway is engageable (co-fold P(binder)=0.92), but a synthetic-lethality play does not target PIK3CA itself; it targets an as-yet-unidentified partner gene whose structure is not in hand. Identifying that partner is an omics/dependency-map task, so it cannot be docked now. Once a partner is nominated it could enter the docking lane; today it is explicitly non-structural.

Docking was only neutral (-6.0/-6.2) and Boltz-2 co-folding refuted engagement (iptm 0.77, P(binder)=0.15). Verdict is ruled-out and it is a conventional small molecule, exactly the class we are pivoting away from. Retire it to free the lane for frontier bets.

Both lanes refuted engagement: docking -2.5 kcal/mol and co-folding iptm 0.25 / P(binder)=0.05. Verdict ruled-out; a trivial small molecule with no frontier value. Retire.

Despite a modest supporting dock (-7.4), co-folding refuted the interaction (iptm 0.90 but P(binder)=0.09, affinity_pred +1.44). Verdict ruled-out; conventional repurposing off-thesis. Retire.

Docking looked favorable (-9.7) but co-folding refuted it (iptm 0.87, P(binder)=0.08), a classic docking false-positive. Verdict ruled-out and conventional. Retire.

Docking supported (-8.1/-10.0) but co-folding strongly refuted (iptm 0.82, P(binder)=0.02). Verdict ruled-out; conventional repurposing. Retire.

Docking supported (-8.8/-9.0) but co-folding refuted engagement (iptm 0.89, P(binder)=0.02). Verdict ruled-out on the second target too; conventional. Retire.

Docking supported (-8.1) but co-folding refuted (iptm 0.95, P(binder)=0.14). Verdict ruled-out; conventional off-thesis. Retire.

Both lanes refuted: docking -4.2 and co-folding iptm 0.87 / P(binder)=0.25. Verdict ruled-out; a nutraceutical, off-thesis. Retire.

Both lanes refuted: docking -4.7 and co-folding iptm 0.58 / P(binder)=0.24. Verdict ruled-out; conventional. Retire.

Docking supported (-7.8) but co-folding refuted (iptm 0.93, P(binder)=0.14). Verdict ruled-out; natural-product repurposing, off-thesis. Retire.