Progress in the Visualization and Mining of Chemical and Target Spaces

Abstract

Chemogenomics is a growing field that aims to integrate the chemical and target spaces. As part of a multi-disciplinary effort to achieve this goal, computational methods initially developed to visualize the chemical space of compound collections and mine single-target structure-activity relationships, are being adapted to visualize and mine complex relationships in chemogenomics data sets. Similarly, the growing evidence that clinical effects are many times due to the interaction of single or multiple drugs with multiple targets, is encouraging the development of novel methodologies that are integrated in multi-target drug discovery endeavors. Herein we review advances in the development and application of approaches to generate visual representations of chemical space with particular emphasis on methods that aim to explore and uncover relationships between chemical and target spaces. Also, progress in the data mining of the structure-activity relationships of sets of compounds screened across multiple targets are discussed in light of the concept of activity landscape modeling.

Exploring Chemical Space with Machine Learning

“…Our main approach to visualize chemical space [43][44][45] consists in generating interactive 2D or 3D maps featuring clouds of color-coded points, each point representing a molecule whose structure is made visible by pointing to it, and whose color encodes a particular property such as the similarity to a reference molecule or a molecular property such as size or polarity. The coordinates of each point on the map are calculated by applying dimensionality reduction to data points in a high-dimensional mathematical space defined by a molecular fingerprint.…”

Progress in the Analysis of Multiple Activity Profile of Screening Data Using Computational Approaches

“…As reviewed before [Rognan, ], target fishing leads to the generation of ligand profiling or the global pharmacological profile. In light of the concept of chemogenomics, target fishing (as well as virtual screening) has emerged as group of computational approaches to try to fill in, as much as possible, the relationships between chemical and target spaces [Rognan, ; Medina‐Franco and Aguayo‐Ortiz, ; Medina‐Franco et al., ]. Same as virtual screening, target fishing can be carried out using structure‐based methods such as inverse docking, ligand‐based approaches like similarity searching [Nettles et al., ] or both.…”

“…A common approach to represent target relationships is using network analysis. Ligand–target networks are mathematical models where nodes represent ligands and targets and the edge linking two nodes represents a cross‐linking interaction e.g., IC 50 or affinity above a predefined threshold [Vogt and Mestres, ; Medina‐Franco and Aguayo‐Ortiz, ]. However, analysis of ligand–target relationships considering polypharmacologic interactions is not a straightforward task due to the many different targets that can be associated for each ligand.…”

Populating Chemical Space with Peptides using a Genetic Algorithm

“…Nevertheless, dimensionality reduction methods enable to represent these spaces as 2D or 3D-maps and lead to a geographical understanding of molecular diversity because molecules with similar properties are found close to one another. [5][6][7][8][9][10][11][12][13] The pertinence of this approach is evidenced by the fact that simple nearest neighbor searches in chemical spaces defined by high dimensional molecular fingerprints often perform as well or even better in virtual screening and target prediction benchmarks than more complex machine learning algorithms. [14][15][16][17][18][19] Given a compound of interest, one would often like to generate new molecules at the same location in chemical space.…”