With the consolidation of deep learning in drug discovery, several novel algorithms for learning molecular representations have been proposed. Despite the interest of the community in developing new methods for learning molecular embeddings and their theoretical benefits, comparing molecular embeddings with each other and with traditional representations is not straightforward, which in turn hinders the process of choosing a suitable representation for Quantitative Structure-Activity Relationship (QSAR) modeling. A reason behind this issue is the difficulty of conducting a fair and thorough comparison of the different existing embedding approaches, which requires numerous experiments on various datasets and training scenarios. To close this gap, we reviewed the literature on methods for molecular embeddings and reproduced three unsupervised and two supervised molecular embedding techniques recently proposed in the literature. We compared these five methods concerning their performance in QSAR scenarios using different classification and regression datasets. We also compared these representations to traditional molecular representations, namely molecular descriptors and fingerprints. As opposed to the expected outcome, our experimental setup consisting of over $25 000$ trained models and statistical tests revealed that the predictive performance using molecular embeddings did not significantly surpass that of traditional representations. Although supervised embeddings yielded competitive results compared with those using traditional molecular representations, unsupervised embeddings tended to perform worse than traditional representations. Our results highlight the need for conducting a careful comparison and analysis of the different embedding techniques prior to using them in drug design tasks and motivate a discussion about the potential of molecular embeddings in computer-aided drug design.

In the modern drug discovery process, medicinal chemists deal with the complexity of analysis of large ensembles of candidate molecules. Computational tools, such as dimensionality reduction (DR) and classification, are commonly used to efficiently process the multidimensional space of features. These underlying calculations often hinder interpretability of results and prevent experts from assessing the impact of individual molecular features on the resulting representations. To provide a solution for scrutinizing such complex data, we introduce ChemVA, an interactive application for the visual exploration of large molecular ensembles and their features. Our tool consists of multiple coordinated views: Hexagonal view, Detail view, 3D view, Table view, and a newly proposed Difference view designed for the comparison of DR projections. These views display DR projections combined with biological activity, selected molecular features, and confidence scores for each of these projections. This conjunction of views allows the user to drill down through the dataset and to efficiently select candidate compounds. Our approach was evaluated on two case studies of finding structurally similar ligands with similar binding affinity to a target protein, as well as on an external qualitative evaluation. The results suggest that our system allows effective visual inspection and comparison of different high-dimensional molecular representations. Furthermore, ChemVA assists in the identification of candidate compounds while providing information on the certainty behind different molecular representations.