Methodology
A narrow rules engine with explicit uncertainty.
v1 is not a thermodynamic simulator. It is a deterministic research-planning model for a curated peptide subset, combining formulation rules, sequence-derived descriptors, and conservative uncertainty penalties.
Evidence-aware by design
The model differentiates between curated entries with broader literature support and entries that still depend on vendor-only or partial metadata.
Condition-specific inputs
Solvent class, preferred pH windows, concentration, light sensitivity, temperature, and freeze-thaw stress all shift the estimate.
Validation-first outputs
Each result points toward the pH, appearance, RP-HPLC, and LC-MS work most likely to confirm or challenge it.
Stage 1
Eligibility gate
Unsupported dosage forms, missing identity, metal-complex systems, or undefined solvent assumptions are rejected before scoring.
Stage 2
Molecular feature derivation
Each curated entry contributes pI, pKa proximity, preferred pH windows, hydropathy, aromaticity, liability burden, and sequence-coverage confidence.
Stage 3
Mixture-state estimation
The engine estimates final pH, concentration, ionic-strength pressure, and solvent-alignment penalties from solvent class and baked-in vendor vial-handling assumptions.
Stage 4
Risk channel scoring
Precipitation, chemical degradation, formulation mismatch, and uncertainty are scored separately before the overall result is assigned.
Stage 5
Hold-time estimate
State-specific baselines are penalized conservatively and presented as research hold time, not shelf life or expiry.
Stage 6
Explanation layer
Every result surfaces top drivers, assumptions, warnings, and required validation assays in plain language.