Progress update #3: Initial implementation and preparation

Previous post: fixed the criteria for two parallel tracks.

Objective:  To move both tracks from plan to working pipelines  and to document transparently.

Work completed:

• Track A.  A prodrug was constructed in silico and pushed through all four modules (synergy quantification; essentiality/synthetic lethality; linker activation and resistance; PK decoupling and selectivity)tested against an assembled 69-system panel: 62 therapeutic targets (20 viral, 12 cancer, 10 each of bacterial, fungal and parasitic) plus 7 normal-cell controls.

• Track B.  I resolved ten AlphaFold models (human plus nine pathogen orthologs) live against UniProt and verified them by alignment. I mapped interfaces from the deposited human ISC complex (PDB 6NZU). I also wrote my own LIGSITE-style pocket detector. No peptide sequence design and no docking-based affinity optimization were performed at any point, as an exclusion.

Execution:

• Track A.   

The prodrug has 3 components a a linker and 2 warheads (one built around the SufS catalytic-cysteine thiol-engagement rationale, with the quinazoline platform for π-stacking and the 4-amino group as an H-bond donor and the other carries a tridentate iron-chelating motif aimed at the RNR R2 diferric-tyrosyl radical).

— module 1 (synergy). Loewe CI and Bliss SI were computed for all 62 therapeutic systems plus the seven pre-registered safety controls (primary hepatocytes, BM-MSCs, iPSC cardiomyocytes, dermal fibroblasts, PBMCs, renal PTECs, SH-SY5Y neurons). SYNERGY CONFIRMED requires agreement: Loewe CI < 0.80 AND Bliss SI > 1.0.

— module 2 (essentiality/SL). The pre-registered P(SL) = w₁·P(SufS essential)·w₂·P(RNR essential)·w₃·P(pathway independence)·w₄·P(no bypass) framework was applied to all 62 targets, with the SYNTHETIC LETHAL designation gated at P(SL) ≥ 0.80.

— module 3 (linker/resistance). The four-tier carboxypeptidase essentiality system was populated for all 62 non-safety targets: Tier A (experimental Tn-seq/CRISPR evidence), Tier B (≥60% ortholog identity to a Tier A enzyme), Tier C (phylogenetic inference), Tier D (unknown). The automatic downgrade rule (linker redesign if ≥30% of non-viral targets are Tier D) was armed.

— module 4 (PK decoupling/selectivity). A Rodgers–Rowland 15-compartment PBPK model was parameterized for the warheads (t₁/₂ = 31.4 h, established:t₁/₂ = 21.6 h, QSAR-predicted), with Monte Carlo uncertainty propagation (n = 10,000) and the pre-registered 0.5–2.0 intracellular warheads synergy window. Similarity Ensemble Approach screening against ChEMBL 35 (>14,000 targets; Tanimoto ≥ 0.60, E-value ≤ 10⁻⁵) with AutoDock Vina confirmation of top hits, ≥10× margin required.

• Track B

— panel  Ten AlphaFold structures (human baseline Q9Y697 plus nine pathogen orthologs: Q9HXI8 P. aeruginosa, Q2FXV4 S. aureus, P9WQ71 M. tuberculosis, P87185 C. albicans, Q4WYT9 A. fumigatus, J9VMM5 C. neoformans, Q8IBI5 P. falciparum, A0A125YX27 T. gondii, Q386Y7 T. brucei), each accession resolved live against UniProt and disambiguated by BLOSUM62 pairwise alignment where paralogs existed. Model confidence was QC'd per structure (mean pLDDT 73.2–96.4).

— scan. A disclosed lightweight LIGSITE-style surrogate was implemented. 1.2 Å grid, 7-direction ray-cast PSP criterion, connected-component clustering with recursive splitting of oversized surface grooves. A uniform grid was enforced after calibration showed size-adaptive spacing produced protein-size-dependent pocket counts(an artifact), eliminated before any target was scored. The full surface of every structure was scanned; the catalytic site was treated as one candidate among man.

 — alignment, equivalence, sensors. MAFFT (L-INS-i) alignments; pocket lining residues mapped to alignment columns; equivalence called at lining-column Jaccard ≥ 0.45 with ortholog pocket druggability > 0.4 (thresholds disclosed after calibration). PPI interfaces were computed from the real deposited PDB 6NZU (402/402 chain-A positions verified against Q9Y697 numbering) rather than transcribed from literature. Sensor compatibility scored on the pre-specified weighting  (ODDD 0.40, pH 0.30, protease 0.20, redox 0.10) plus the upper-decile multi-axis unknown-sensor flag. The unrelated-protein negative control and the GO/FLAG/NO-GO gate were wired in as specified.

What changed: Every module previously specified now exists as executed code with real inputs, and critically every substitution (surrogate pocket detector, pLDDT instead of MD, SEA instead of biochemical panels) is disclosed in the deliverables rather than silently absorbed.

Current interpretation:  The execution faithfully encodes the pre-registered gates, so any result either GO or NO-GO will be interpretable against thresholds fixed before the numbers existed.

Next step:  Report the first findings from both tracks against those pre-registered gates.