Pre-Registered De-Novo Peptide Design for HCT-116 Colorectal Carcinoma: Methodology, Predicted IC50, and Wet-Lab Validation Protocol (v1.0)

### Background and scope This pre-registration documents the methodology, predicted outcomes, and wet-lab validation protocol for three de-novo therapeutic peptide candidates designed to target **HCT-116 (human colorectal carcinoma, ATCC CCL-247, KRAS G13D mutant)**. The three candidates span three orthogonal oncology anchors — cytotoxic (membrane-anchoring tumor-homing family), kinase-binding (HER-family ATP-pocket via reduced-HBD backbone N-methylation), and DNA-binding (NLS-mediated nuclear translocation via N-terminal lipidation) — enabling a single HCT-116 plate to test multiple mechanism hypotheses simultaneously with minimal experimental n. ### Methodology Peptide candidates were generated by an ESM-2 (Meta AI, 650M-parameter protein language model) Gibbs sampling pipeline seeded on literature-validated motif families, with deterministic reproducibility guaranteed across both Python and PyTorch RNG state. Each candidate was evaluated by a 5-axis integrated readiness score combining (1) **permeability** (Lipinski Ro5 + Veber + Caco-2 prediction, with mechanism-aware overrides for non-passive-diffusion classes such as cyclosporin-class N-methylated peptides and fatty-acid-anchored lipopeptides), (2) **synthesis risk** (sequence-level pattern detection for Trp oxidation, Asn-Gly aspartimide, Pro clustering, multi-Cys disulfide risk), (3) **HCT-116 IC50 prediction** by k=20 nearest-neighbor weighted log-IC50 against a 192-compound reference pool (DepMap CTRP × DrugBank + PubChem peptide bioactivities), (4) **novelty** (Morgan fingerprint Tanimoto similarity against a ~22K-peptide internal corpus from SATPdb / APD3 / CPPsite / THPdb), and (5) **seed-motif preservation** (longest common substring of seed motif within ESM-2 resampled sequence). Permeability-mechanism-specific multiplicative penalty factors (× 5 for membrane anchoring, × 3 for reduced HBD, × 1 for active CPP transport, × 10–50 for passive diffusion by Lipinski-violation count) are applied to the raw IC50 prediction, with resulting 95% confidence intervals and reliability flags (high / medium / low) reported per candidate. ### Predicted outcomes The three candidates are predicted to have HCT-116 IC50 values in the single-digit to low-double-digit micromolar range — specifically **6.4 μM**, **9.3 μM**, and **11.65 μM** with 95% confidence intervals of 1.33–3.42, 0.17–20.76, and 1.67–3.24 μM respectively (all classified as **CONDITIONAL grade B** under the Option B global-pharma honest standard, with explicit **medium** or **low** reliability flags acknowledging the peptide-vs-small-molecule reference pool mismatch). Integrated readiness scores are 0.624, 0.725, and 0.684. Wet-lab validation will follow a 5-point CCK-8 cytotoxicity assay protocol (concentrations 0.1 / 1 / 10 / 100 / 1,000 μM, 72-hour incubation, Melittin 10 μM positive control, DMSO 0.1% v/v negative control, GraphPad Prism 4-parameter logistic fit, n = 3 technical × ≥ 2 biological replicates). ### Pre-registered hypotheses Four falsifiable hypotheses are formally registered with explicit Bayesian decision criteria: (H1) at least one candidate measures IC50 ≤ 30 μM (~95% prior probability); (H2) the two membrane-anchoring candidates cluster on (IC50, Hill slope) distinguishably from the reduced-HBD candidate (~55% prior); (H3) per-candidate predicted IC50 falls within ± 10× of measured (~60% prior for 2+ of 3); (H4) the medium-vs-low reliability flag is itself calibrated, i.e., the low-reliability candidate shows the largest prediction-vs-measurement deviation (~70% prior). Each hypothesis has explicit falsification criteria and per-outcome implications for v8 system rebuild direction. ### Disclosure boundary (Composition-of-matter IP protection) This release documents the methodology, predicted IC50 values with confidence intervals and reliability flags, all evaluation formulas and thresholds, the wet-lab protocol, four falsifiable hypotheses, and nine known systematic limitations. **Specific peptide sequences, SMILES strings, backbone N-methylation positions, and seed-motif source publications are intentionally withheld** pending composition-of-matter patent filing. A subsequent release (v2) will disclose these following wet-lab result acquisition and IP securement (planned 2026 Q3). This pre-registration follows OSF (Open Science Framework) and ClinicalTrials.gov standards for protocol pre-registration. The explicit timestamping of methodology and predictions before measurement strengthens the rigor and reproducibility of the subsequent v2 results disclosure. ### Files in this release `README.md` (overview and citation) · `01_methodology.md` (pipeline architecture: ESM-2 + modifier + 5-axis readiness) · `02_oncology_anchors.md` (3 oncology anchors at class level) · `03_candidates_anonymized.md` (per-candidate predicted IC50, CI, reliability, readiness) · `04_wetlab_protocol.md` (HCT-116 5-point CCK-8 protocol with controls and statistics) · `05_prereg_hypotheses.md` (4 falsifiable hypotheses with Bayesian decision matrix) · `06_honest_limitations.md` (9 systematic limitations with mitigation paths) · `07_evaluation_metrics.md` (all formulas, thresholds, and weights) · `LICENSE` (CC-BY-NC-4.0) · `.zenodo.json` (machine-readable metadata). ### License and use This pre-registration is released under **CC-BY-NC-4.0 (Attribution + NonCommercial)**. Academic use, derivative works, and reproducibility verification are encouraged. Commercial use requires a separate license agreement. The withheld composition-of-matter information is not licensed under this release. ---