BioDesign-2026: Ordinal MIC Prediction for Antimicrobial Peptide Discovery Across WHO-Priority Pathogens

BioDesign-2026 is an end-to-end computational platform for de novo antimicrobial peptide (AMP) discovery, purpose-built for WHO-priority drug-resistant pathogens. To our knowledge, no existing platform combines quantitative ordinal MIC prediction, organism-specific modelling, and integrated mammalian safety profiling within a single discovery pipeline. Performance metrics reflect v20.0 architecture (4-class ordinal classifier). Earlier versions used a binary classification framework; metrics are not directly comparable across architectures. The problem it solves. Existing AMP classifiers output binary active/inactive labels, discarding the quantitative structure of MIC data and conflating peptides with MIC=1 µg/mL and MIC=64 µg/mL under the same label. BioDesign-2026 replaces binary classification with 4-class ordinal prediction (INACTIVE / WEAK / ACTIVE / POTENT), providing clinically actionable output across 12 organisms — including all 6 WHO Critical-priority Gram-negative pathogens. Validated performance on independent holdout data (n=3,528, strict sequence-level split, 0.00% train/test overlap): Macro MCC = 0.580 | Macro AUC = 0.851 | Macro Accuracy = 80.3% Per-organism AUC range: 0.806–0.894 Outperforms four established tools on an independent benchmark (n=150 peptides, 3 organisms, 50% balanced): Tool MCC AUC BioDesign-2026 (this work) +0.409 0.784 AMPir +0.048 0.524 AI4AMP −0.140 0.430 AMP Scanner v2 — 0.477 McNemar test p