Model Performance Series
June 2026
Astra AI on Transporters
Membrane transporters move ions, nutrients, and drugs across the lipid bilayer — among the most under-exploited drug-target classes, and among the hardest to characterize.
Çağlar Bozkurt, Aniruddh Goteti · Orbion GmbH · Benchmarked on 6,203 Transporters
What We Found
Three Results That Matter
Each section of the whitepaper reports the headline performance for one prediction area, its known weaknesses, and where we recommend it for production use.
97% Recognition Across the Canonical Families
SLC carriers 97%; ABC transporters and aquaporins 100% recognized as transporters. Topology holds at AUROC 0.94–0.98 across the groups.
F1 Up to 0.88 on PTM Sites
Across 39 modification classes, strongest on disulfide bonds (0.88) and N-linked glycosylation (0.83) — both critical to multi-pass carrier function.
ρ 0.59 ΔTm Ranking on SERT
On serotonin-transporter mutations the model ranks by predicted ΔTm at ρ 0.59, MAE 2.13 °C, with 81% directional accuracy on strong stabilizers.
Every number above is reproduced in the whitepaper against public references — Swiss-Prot annotation, PDB co-crystal contacts, and curated experimental thermal-shift data. We report where the models are weak as plainly as where they are strong.