Best Practices in Docking and Activity Prediction

Abstract

During the last decade we witnessed how computational docking methods became a crucial tool in the search for new drug candidates. The central dogma of small molecule docking is that compounds that dock correctly into the receptor are more likely to display biological activity than those that do not dock. This dogma, however, possesses multiple twists and turns that may not be obvious to novice dockers. The first premise is that the compounds must dock; this implies: (i) availability of data, (ii) realistic representation of the chemical entities in a form that can be understood by the computer and the software, and, (iii) exhaustive sampling of the protein-ligand conformational space. The second premise is that, after the sampling, all docking solutions must be ranked correctly with a score representing the physico-chemical foundations of binding. The third premise is that correctness must be defined unambiguously, usually by comparison with static experimental data (or lack thereof). Each of these premises involves some degree of simplification of reality, and overall loss in the accuracy of the docking predictions.\n\nIn this chapter we will revise our latest experiences in receptor-based docking when dealing with all three above-mentioned issues. First, we will explain the theoretical foundation of ICM docking, along with a brief explanation on how we measure performance. Second, we will contextualize ICM by showing its performance in single and multiple receptor conformation schemes with the Directory of Useful Decoys (DUD) and the Pocketome. Third, we will describe which strategies we are using to represent protein plasticity, like using multiple crystallographic structures or Monte Carlo (MC) and Normal Mode Analysis (NMA) sampling methods, emphasizing how to overcome the associated pitfalls (e.g., increased number of false positives). In the last section, we will describe ALiBERO, a new tool that is helping us to improve the discriminative power of X-ray structures and homology models in screening campaigns.