2026/05/12 by Insan Kang · 1 voice
Biochemistry, Genetics and Molecular Biology · #RNA and protein synthesis mechanisms #Chemical Synthesis and Analysis #Protein Structure and Dynamics
paper · doi:10.26434/chemrxiv.15002948/v2
openalex publication_date 2026/05/12 · openalex created_date 2026/05/13 · openalex updated_date 2026/07/15
Non-canonical amino acid (ncAA) substitutions are central to peptide-drug development, yet quantitative in silico ranking of ncAA variants remains constrained by force-field junction parameters at the canonical / non-canonical boundary, charge-derivation provenance, and the limited reproducibility of MM-PBSA / MM-GBSA scoring at small mutational scales. Khoury et al. (2014) reported a comprehensive same-system MM-GBSA / experimental correlation of R² = 0.388 over 147 ncAAs, defining a defensible field ceiling that newer methods only partially close. We present UPDD (Universal Peptide Drug Discovery), a five-stage iterative pipeline whose differentiating contribution is its Stage-4 evaluation framework for ncAA-modified peptide complexes (ncAA parameterisation → restrained MD → QM/MM single-point → MM-PBSA scoring), integrated with established generation tools (RFdiffusion + ProteinMPNN + AlphaFold2). We position UPDD at Capability Level 1: a decision-support tool for ncAA variant ranking, not a quantitative experimental predictor. Three methodological contributions are documented: (1) a σbtwn / σw ensemble-quality decomposition with a single-metric Convergence Index (CI = σbtwn / |⟨⟨ΔG⟩⟩|) carrying a dual interpretation (binding-mode classifier and automated quality-control auditor); (2) a branched ΔΔG architecture with bit-identical wild-type control (intervention isolation verified at floating-point precision); and (3) a multi-layer verification protocol empirically validated by four sequential audit layers — closed-shell charge bias, periodic-boundary serialisation, intra-residue bond integrity for HETATM ncAAs, and source-versus-test consistency. Across six ncAA peptide-target families (1EBP / 1YCR / 2QKH / 2QKI / 3IOL / 7TL8), the framework demonstrates sign-direction reproducibility on a validated subset, six-of-six Definition-3 readiness, and three biophysical detachment events distinguished from periodic-image artefacts. We acknowledge the Khoury R² = 0.388 ceiling, position the next charge-methodology axis (Khoury full RESP) as the planned v0.8 work, and decline universal quantitative-match claims. An iterative active-learning loop with wet-lab Kd feedback is operationally specified, and closed-loop calibration remains the v0.8 / v1.0 milestone.