Automated Inference of Chemical Discriminants of Biological Activity

Abstract

Ligand-based virtual screening has become a standard technique for the efficient discovery of bioactive small molecules. Following assays to determine the activity of compounds selected by virtual screening, or other approaches in which dozens to thousands of molecules have been tested, machine learning techniques make it straightforward to discover the patterns of chemical groups that correlate with the desired biological activity. Defining the chemical features that generate activity can be used to guide the selection of molecules for subsequent rounds of screening and assaying, as well as help design new, more active molecules for organic synthesis.The quantitative structure-activity relationship machine learning protocols we describe here, using decision trees, random forests, and sequential feature selection, take as input the chemical structure of a single, known active small molecule (e.g., an inhibitor, agonist, or substrate) for comparison with the structure of each tested molecule. Knowledge of the atomic structure of the protein target and its interactions with the active compound are not required. These protocols can be modified and applied to any data set that consists of a series of measured structural, chemical, or other features for each tested molecule, along with the experimentally measured value of the response variable you would like to predict or optimize for your project, for instance, inhibitory activity in a biological assay or ΔG. To illustrate the use of different machine learning algorithms, we step through the analysis of a dataset of inhibitor candidates from virtual screening that were tested recently for their ability to inhibit GPCR-mediated signaling in a vertebrate.

Comparison Study of Computational Prediction Tools for Drug-Target Binding Affinities

“…AI/ML and statistical analysis approaches have been applied across different stages of the drug development and design pipelines (Lima et al, 2016) including target discovery (Ferrero et al, 2017), drug discovery (Hutter, 2009; Raschka et al, 2018; Vamathevan et al, 2019), multi-target drug combination prediction (Tang et al, 2014; Vakil and Trappe, 2019), and drug safety assessment (Raies and Bajic, 2016, 2018; Lu et al, 2018). AI/ML approaches are generally either feature-based or similarity-based.…”

Machine learning and AI-based approaches for bioactive ligand discovery and GPCR-ligand recognition

“…However, the virtual screening data combined with the activity measurements collected via experimental assay data offer excellent opportunities to utilize machine learning for developing custom activity classifiers and conducting quantitative structure–activity relationship (QSAR) analyses. For instance, in a follow-up study, the researchers described how machine learning techniques were used to identify the chemical features that correlated with bioactivity [ 25 ]. To facilitate this analysis, the researchers converted functional group matches between database compounds and the receptor agonist into binary feature vectors.…”

“…To illustrate these supervised learning concepts in the context of GPCR bioactive ligand discovery, we revisit the SLOR1 receptor signaling inhibitor projects [ 24 , 25 ] introduced in Section 1.2.1 . Suppose the goal is to train a classification model to predict whether a (new) candidate molecule is active against SLOR1.…”

“…Next, we have to decide about the feature representation of the molecules that are presented to the machine classifier along with the target labels. For example, we may use the functional group matching vectors based on 3D volumetric overlays with a known receptor agonist as described in Section 1.2.1 and [ 25 ]. Before we start training the classifier, we divide the dataset into a separate training and test set, where the training set is used to fit the model.…”

“…For example, in the SLOR1 context discussed in Sections 1.2.1 and 1.3 , the researchers were able to identify a potent inhibitor of GPCR pheromone receptor signaling using a machine learning-aided virtual screening approach that provided structure–activity information [ 24 ]. Furthermore, by combining machine learning with feature selection algorithms, the researchers were able to identify sulfate groups on the tail end of the candidate ligands that are crucial for bioactivity in SLOR1 GPCR signaling [ 25 ]. While the methods described in [ 25 ] were applied to summarize structure–activity relationships of 3D-structural overlays of small molecules that were prioritized for experimental bioassays, the same techniques can be used for different types of data, such as hydrogen bonding patterns [ 62 ] or ligand-receptor docking poses [ 63 ].…”