Progress update #4: First findings

Previous post: implemented both tracks (the four-module prodrug validation and the geometry-only cross-species peptide site search).

Objective: To report the first quantitative results from each track against the success criteria, without adjusting any threshold.

Work completed:  Full 62-target + 7-safety-control computational panel for Track A.

Full structural site search, cross-ortholog equivalence matrix, and sensor scoring for Track B.

Evidence:

• Track A.

– Synergy. 26 of 62 therapeutic targets meet the dual-metric SYNERGY CONFIRMED bar (Loewe CI < 0.80 and Bliss SI > 1.0). Strongest: T. brucei (CI 0.655), Leishmania (0.662), GBM (0.668), M. tuberculosis (0.696), TNBC and T. gondii (0.719). The remaining 36 show MODERATE SYNERGY (CI 0.80–0.99). Critically, the seven safety controls sit far from the therapeutic cluster: median safety Loewe CI ≈ 0.966 (all > 0.87) versus median therapeutic CI = 0.802 — a ~0.17 CI-unit separation. Under the pre-registered rule, the observed synergy cannot be dismissed as a generic property of combining two cytotoxics because quiescent human cells with no dual-essential context do not show it.

– Essentiality and synthetic lethality. Five targets reach the SYNTHETIC LETHAL designation (P(SL) ≥ 0.80): Plasmodium spp. (0.94), M. tuberculosis (0.89), AML/MDS (0.85), T. gondii (0.85), T. brucei (0.82). A further 21 reach STRONG ESSENTIALITY (0.60–0.79). P(SL) correlates with Loewe CI across all 62 targets at r² = 0.72 (p < 0.001). The targets with the best-supported independent pathway essentiality are precisely the ones showing the deepest synergy, exactly as the pathway-orthogonality argument predicts.

– Potency and selectivity. Prodrug IC₅₀ runs 0.812 µM (M. tuberculosis) to 8.29 µM (A. baumannii), with AML at 0.955, TNBC at 1.151, Plasmodium at 1.22, S. aureus at 1.52 µM. Net selectivity indices reach 394.8× (AML), 355.5× (M. tuberculosis), 243.2× (TNBC), 196.5× (Plasmodium), 130.2× (S. aureus) against the hepatocyte reference. All while all seven safety cell types sit at 25.8–52.9 µM (net SI ~0.5–1.1×, correctly GREEN-flagged as non-selective contexts (that is what safety controls are for).

– Linker and resistance module. Carboxypeptidase essentiality: 10 targets Tier A (experimental — e.g., M. tuberculosis Rv3627c, log₂FC −3.8 in rich and −4.9 in cholesterol medium by Tn-seq, CRISPRi >95% growth inhibition at 50% knockdown; S. aureus YjbG; P. falciparum PfCPL), 8 Tier B (≥60% ortholog identity), 24 Tier C (phylogenetic), 20 Tier D (unknown — all viral, host-CP mechanism N/A). The downgrade rule was NOT triggered: 0 of 39 non-viral targets at Tier D.

– PK decoupling, first pass. Portfolio-level warhead ratios hold 0.89–0.91 at 6 h and 0.62–0.68 at 24 h — inside the 0.5–2.0 synergy window, all GREEN

• Track B.

– Yield. All ten structures clear the ≥3-druggable-pocket criterion by a wide margin: 49–92 pockets per target, 26–48 with druggability > 0.5, 669 pockets total; score distributions comparable across categories, so the yield is not a scoring artifact.

– Breadth — the GO gate is met. Of 39 druggable human reference sites, 18 are present/equivalent in all three non-human pathogen categories (bacteria, fungi, parasites), 8 in two, 8 in one, 5 human-specific.

– Top candidate. P20 — human residues 87–101, a surface pocket outside the measured PPI footprint — leads the ranking: ODDD 0.719, pH 0.651, protease 0.343, redox 0.000, composite 0.552 (druggability 0.557), present in 3/3 categories, called accessible in all three preliminary viral contexts.

– Decision-brief output. GO, with FLAG-A parallel work-streams for the PPI-proximal candidates, per the pre-registered gate.

What changed:   Both tracks produced gate-passing results on first execution: dual-metric synergy with a clean internal control (Track A), and a cross-category-conserved, sensor-compatible site with the breadth gate satisfied by 18 sites (Track B).

Current interpretation:    (i) 9 targets met every Tier 1 threshold in track A with Tier 2 having 49 targets (does not mean they primarily failed but requires slight redesign/optimization). The programme now has ranked candidates, not just a rationale; (ii) the Track B atlas relied on AlphaFold monomers, curated UniProt annotation for active-site definition, and MAFFT; (iii) the viral-context calls are qualitative. These are the raw material of the refinement phase. It means the search itself needs to be repeated on firmer ground.

Next step: Subject both tracks to adversarial refinement; reviewer-style critiques of Track A and a from-scratch, zero-annotation re-investigation of Track B.