Deep Learning for Virtual Screening: Five Reasons to Use ROC Cost Functions
V. Golkov, A. Becker, D. T. Plop, D. Cuturilo, N. Davoudi, J. Mendenhall, R. Moretti, J. Meiler, D. Cremers
This paper argues for using ROC cost functions in deep learning for virtual screening to address class imbalance and decision threshold challenges, showing improved performance on drug discovery datasets.
bioRxiv · Jun 25, 2020 · PreprintPharmacology & Toxicology