SPADE Algorithm Cuts Drug Discovery Tests by 60%
A new algorithm called SPADE (Sparse Data Exploration) can find 10 high-quality drug candidates in just 40 tests on average, outperforming deep learning and Bayesian optimization methods. The algorithm achieves median improvements of 7%-32% in sample efficiency and is 10 times faster than its closest competitor at scoring candidate drugs. SPADE is designed for novel proteins with no prior data, addressing a key bottleneck in drug discovery where fewer than 5% of candidate ligands pass early stages. The dataset and code are publicly available.
Key facts
- SPADE requires only 40 tests on average to find 10 high-quality ligands
- SPADE outperforms deep learning and Bayesian optimization methods on more proteins
- Median improvements of 7%-32% in sample efficiency
- SPADE is 10x faster than its closest competitor at scoring candidate drugs
- Designed for novel proteins with no prior data
- Fewer than 5% of candidate ligands pass early stages of drug discovery
- Dataset and code available at anonymous.4open.science/r/SPADE_Fast_Drug_D
- Published on arXiv with ID 2605.05370
Entities
Institutions
- arXiv