Unlocking Scalable Ligand Residence Time Predictions with Koffee Unbinding Kinetics Simulations

Abstract

A great number of drug discovery programs fail due to poor in vivo efficacy and ADMET liabilities. On- and off-target ligand residence times can act as important drivers of these problems. While modern experimental techniques have made measuring compound kinetics data more routine, there is a lack of accurate, high-throughput simulation techniques to guide compound prioritization by residence time. In this work, we introduce Koffee™ Unbinding Kinetics as a solution to the hitherto unanswered problem of scalable ligand-protein residence time prediction. By bypassing conventional approaches based on molecular dynamics simulations, Koffee Unbinding Kinetics performs physics-based residence time screening at the atomistic level in ≈1 GPU minute per complex using inexpensive hardware, a speed-up of at least 3-5 orders of magnitude compared to current state-of-the-art simulation approaches. Koffee Unbinding Kinetics can enhance compound selection to mitigate costly future program failures by adding fast, predictive residence time simulations to early-stage computational drug discovery pipelines.