Gaussian processes regression for cyclodextrin host-guest binding prediction

Abstract

Machine Learning (ML) techniques are becoming an integral part of rational drug design and discovery. Data-driven modeling regularly outperforms physics-based models for predicting molecular binding affinities, placing ML as a promising tool. Cyclodextrins are nano-cages used to improve the delivery of insoluble or toxic drugs. Due to chemical similarity to proteins, ML approaches could vastly profit to improve affinity prediction and enhance their carriable drug portfolio. Here we evaluate the performance of three well-known ML methods -Support Vector Regression (SVR), Gaussian Process Regression (GPR), and eXtreme Gradient Boosting (XGB) -to predict the binding affinity of cyclodextrin and known ligands. We perform hyperparameter tuning through Random Search. The results were compatible with the presented literature. We were able to increase our previous prediction performance and present a GPR model to adjust to the data (R 2 = 0.803) with low prediction errors (RMSE = 1.811 kJ/mol and MAE = 1.201 kJ/mol).

Primary vs secondary: Directionalized guest coordination in β-cyclodextrin derivatives

“…Among so many macrocyclic receptors, e.g., cyclodextrin (CD), calixarene, cucurbit[n]uril and pillar[n]arene, the most commonly employed macrocycle is CD, due to their readily availability, low cost and high solubility. [1][2][3][4] Cyclodextrins (CDs) are a family of cyclic oligosaccharides widely used in pharmaceutical and chemical industries. The CD derivatives share the macrocyclic glucose backbone, which forms a hydrophobic cavity and a hydrophilic exterior.…”

Host Dynamics under General-Purpose Force Fields

“…Cyclodextrins are probably the most commonly applied macrocycles due to their readily availability, low cost, and tunable solubility [ 62 , 63 , 64 ]. Their glucose units have many donors and acceptors for hydrogen bonds and hold a hydrophobic cavity encapsulating drug-like molecules.…”

Curated dataset of association constants between a cyclodextrin and a guest for machine learning

“…Indeed, the application of machine learning to chemistry has increased significantly in recent years [6,7]. These methods are also becoming very popular for the determination of the CD/guest association constant [8][9][10][11]. Their first requirement is the availability of a quality dataset.…”

“…However, in some publications, the dataset used is not publicly available, the experimental conditions are not presented or the data format not easily reusable for other machine learning studies. For example, (i) 233 data are available but only for β-CD or sulfobutylether-β-CD [8] (ii) 1320 data are used including 8 different CDs but the dataset is not available [9], (iii) 1654 data are available included 16 different CDs acquired from the Cyclodextrin knowledge base (8534 collected data) unfortunately no longer available [12] and (iv) 280 data are available from the BindingDB community [11,13].…”

“…In this data paper, we propose a curated dataset of association constants. The data come from different articles/sources from the years 1963 to 2021 [10,11,[14][15][16][17][18][19]. For each CD/guest pair (stoichiometry 1/1), the values of association constant K (or Gibbs free energy ΔG with ΔG = -RTln(K)), pH, temperature (T), and the source of data are presented in the data files.…”