Peptide AI / 2026

AMPCliff.

Quantitative definition and benchmarking of activity cliffs in antimicrobial peptides.

Journal of Advanced Research · 80, 287–300

Richard’s author role: Co-corresponding author

Read paper ↗Explore public code ↗Chinese coverage (link supplied by Richard) ↗

01 / The problem

Small peptide changes can produce large antimicrobial-activity changes. Ordinary splits may obscure this activity-cliff challenge.

02 / The core idea

Quantitatively define activity cliffs in antimicrobial peptides and benchmark models with dedicated evaluation splits.

How the idea flows

FIGURE 01 / ORIGINAL CONCEPT SKETCH

Antimicrobial peptide activity records
Define sequence–activity cliffs
Dedicated benchmark splits
Compare models
Inspect cliff-specific failure

Original concept sketch; not a reproduced paper figure.

03 / What the public evidence shows

The publication defines the challenge rather than presenting it as solved. In the reported 33-layer ESM2 −log(MIC) regression setting, Spearman correlation is 0.4669: a concrete sign of remaining difficulty.

Source: AMPCliff ↗

04 / Where the claim stops

The reported correlation is specific to its benchmark and metric, not a high-accuracy therapeutic-design claim. Online publication in 2025 and the 2026 volume describe one paper.

An editorial scope note, not an unverified quotation of author limitations.