Proteome scale drug prioritization from AlphaFold 3 interaction confidence
- 1University at Buffalo
Abstract
AlphaFold 3 (AF3) predicts protein ligand complexes, but the utility of its confidence scores for drug discovery remains unclear. We benchmarked AF3 interaction scores against BANDOCK+, an experimentally informed approach, using approved drugs and human protein libraries. AF3 ipTM achieved slightly higher observed aggregate disease ranking scores in a matched 1,103 protein comparison, with performance varying among individual indications. BANDOCK+ enabled substantially broader protein coverage using a computationally less demanding approach, including a 20,158 protein reference library. These findings highlight the potential of AF3 confidence scores for computational drug discovery and the tradeoffs between predictive performance and scalability.
Related
- CHORUS: A proteome-scale atlas of membrane interaction interfaces constructed using cross-protein language model embeddings from sequences alone
- Retrieval of binding sites across the AlphaFold human proteome using protein language model representations
- AF3 Inspector: In-Browser 3D Model Visualization and Confidence Analytics for AlphaFold 3
- Population-scale drug target and variant prioritization with Varformer
- NuFold RNA Structure Database: Large-Scale RNA Tertiary Structure Prediction with Confidence Metrics
BibTeX
@misc{akabari2026proteome,
title = {{Proteome scale drug prioritization from AlphaFold 3 interaction confidence}},
author = {R. Akabari and W. Mangione and Z. Falls and R. Samudrala},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.19.757941},
url = {https://www.biorxiv.org/content/10.64898/2026.10.19.757941v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Proteome scale drug prioritization from AlphaFold 3 interaction confidence
AU - R. Akabari
AU - W. Mangione
AU - Z. Falls
AU - R. Samudrala
PY - 2026
DA - 2026/10/19
PB - bioRxiv
DO - 10.64898/2026.10.19.757941
UR - https://www.biorxiv.org/content/10.64898/2026.10.19.757941v1
AB - AlphaFold 3 (AF3) predicts protein ligand complexes, but the utility of its confidence scores for drug discovery remains unclear. We benchmarked AF3 interaction scores against BANDOCK+, an experimentally informed approach, using approved drugs and human protein libraries. AF3 ipTM achieved slightly higher observed aggregate disease ranking scores in a matched 1,103 protein comparison, with performance varying among individual indications. BANDOCK+ enabled substantially broader protein coverage using a computationally less demanding approach, including a 20,158 protein reference library. These findings highlight the potential of AF3 confidence scores for computational drug discovery and the tradeoffs between predictive performance and scalability.
ER -