AWS-JHU Antibody Developability Benchmark

First large-scale heterogeneous antibody developability dataset with 50 seed antibodies across 4 structural formats, 42 antigens, 6 developability traits, all wet-lab validated. Supports zero-shot evaluation.

Composite

72.3

Experimental validation

Wet-lab confirmed

Stages

Developmental CandidateLead ID / ADMET

Modalities

biologic

Task types

classificationregression

Size

seed_antibodies: 50
structural_formats: 4
antigens: 42
developability_traits: 6

License

Other

First release

2026-04-14

Last updated

2026-04-14

Official site

→ project page

Leaderboard

→ leaderboard

Dataset

→ dataset

Code / GitHub

→ repository

HuggingFace

→ HF

Paper

AWS-JHU Antibody Developability Benchmark · Jeffrey Gray, AWS AI Labs · 2026 · paper · doi:N/A — paper pending · 0 citations

Flags

none

Experts

—

Groups

—

Hosted by

—

Related benchmarks

FLAb2 (Fitness Landscape for Antibodies 2), BenchBB (Bench-tested Binder Benchmark), AbBiBench (Antibody Binding Benchmark)

Rubric (7-criterion)

rigor

coverage

maintenance

adoption

quality

accessibility

industry_relevance

Notes

Groundbreaking for antibody developability — fills gap where most benchmarks focus on binding only. Wet-lab validated ground truth across diverse formats. Zero-shot evaluation support is novel. AWS backing suggests long maintenance.

← Back to all benchmarks

Compare:

Open comparison →