AI
Mirror-Score: Calibrated, Inference-only Scoring Exposes the Limits of Sequence-compatibility Ranking in D-peptide Design
Publication: cs.AI updates on arXiv.org
Author: Jiada Li
Source published: 30/09/2026, 04:00:00 UTC
Briefing created: 30/09/2026, 04:15:55 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 04:03:25 UTC.
Briefings
Mirror-Score is a calibrated inference-only scoring framework for heterochiral D-peptide/L-protein complexes, replacing raw ProteinMPNN NLL as a valid affinity ranker. It evaluated 31 crystal complexes across four target families, finding that Boltz-2 mirror-space cofolding achieves structure-level leave-one-out Spearman rho = 0.90 for the viral-entry family, whereas NLL failed with rho = 0.18. Cross-family calibration does not transfer at current sample sizes, supporting family-matched calibration as the practical deployment mode. The framework includes a prospective design protocol for antimicrobial-resistance targets LasR and LecB from Pseudomonas aeruginosa.
Key points
- Mirror-Score uses Boltz-2 mirror-space cofolding to achieve structure-level leave-one-out Spearman rho = 0.90 for the viral-entry family, whereas raw ProteinMPNN NLL failed with rho = 0.18.
- The framework supports family-matched calibration as the practical deployment mode since cross-family calibration does not transfer at current sample sizes.
Context and limitations
The benchmark includes only 31 crystal complexes across four target families, with only three independent chemotypes represented in the viral-entry family, limiting generalizability.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 04:15:55 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeifq3b33rddkm3aok3rz4ffp6h4t4hrrk4gxgaihejhtbhzbj5y3sm