01 / The problem
Peptide AI / 2026
PepHarmony.
Sequence–structure contrastive learning for peptide representation.
Neural Networks · 194, 108148
02 / The core idea
Align sequence and structure representations by multi-view contrastive learning; transfer structural information into the sequence encoder.
How the idea flows
FIGURE 01 / ORIGINAL CONCEPT SKETCH
- Paired sequences + structures
- Sequence ∥ structure encoders
- Contrastive alignment in training
- Sequence-only inference
- Representation → downstream task
03 / What the public evidence shows
The public author manuscript reports baseline comparisons and ablations of contrastive loss and ordering. Public code provides affinity, cell-penetration and solubility evaluation resources.
Source: PepHarmony ↗Earlier author manuscript ↗04 / Where the claim stops
The linked author manuscript is an earlier version; the final journal citation is listed above. Representation benchmarks do not demonstrate therapeutic efficacy or deployment.
An editorial scope note, not an unverified quotation of author limitations.