Drug screening resembles finding a needle in a haystack: identifying a few effective inhibitors from a large pool of potential drugs. Large experimental screens are expensive and time-consuming, while virtual screening trades off computational efficiency and experimental correlation. Here we develop a framework that combines molecular dynamics (MD) simulations with active learning. Two components drastically reduce the number of candidates needing experimental testing to less than 20: (1) a target-specific score that evaluates target inhibition and (2) extensive MD simulations to generate a receptor ensemble. The active learning approach reduces the number of compounds requiring experimental testing to less than 10 and cuts computational costs by ∼29-fold. Using this framework, we discovered BMS-262084 as a potent inhibitor of TMPRSS2 (IC50 = 1.82 nM). Cell-based experiments confirmed BMS-262084’s efficacy in blocking entry of various SARS-CoV-2 variants and other coronaviruses. The identified inhibitor holds promise for treating viral and other diseases involving TMPRSS2.

In this work, we propose a novel method that combines in silico prediction of molecular properties such as biological activity or pharmacokinetics with an in silico optimization algorithm, namely Particle Swarm Optimization. Our method takes a starting compound as input and proposes new molecules with more desirable (predicted) properties. It navigates a machine-learned continuous representation of a drug-like chemical space guided by a defined objective function. The objective function combines multiple in silico prediction models, defined desirability ranges and substructure constraints. We demonstrate that our proposed method is able to consistently find more desirable molecules for the studied tasks in relatively short time.

The COVID-19 pandemic continues to pose a substantial threat to human lives and is likely to do so for years to come. Despite the availability of vaccines, searching for efficient small-molecule drugs that are widely available, including in low- and middle-income countries, is an ongoing challenge. In this work, we report the results of a community effort, the “Billion molecules against Covid-19 challenge”, to identify small-molecule inhibitors against SARS-CoV-2 or relevant human receptors. Participating teams used a wide variety of computational methods to screen a minimum of 1 billion virtual molecules against 6 protein targets. Overall, 31 teams participated, and they suggested a total of 639,024 potentially active molecules, which were subsequently ranked to find ‘consensus compounds’. The organizing team coordinated with various contract research organizations (CROs) and collaborating institutions to synthesize and test 878 compounds for activity against proteases (Nsp5, Nsp3, TMPRSS2), nucleocapsid N, RdRP (Nsp12 domain), and (alpha) spike protein S. Overall, 27 potential inhibitors were experimentally confirmed by binding-, cleavage-, and/or viral suppression assays and are presented here. All results are freely available and can be taken further downstream without IP restrictions. Overall, we show the effectiveness of computational techniques, community efforts, and communication across research fields (i.e., protein expression and crystallography, in silico modeling, synthesis and biological assays) to accelerate the early phases of drug discovery.

The COVID‐19 pandemic continues to pose a substantial threat to human lives and is likely to do so for years to come. Despite the availability of vaccines, searching for efficient small‐molecule drugs that are widely available, including in low‐ and middle‐income countries, is an ongoing challenge. In this work, we report the results of an open science community effort, the “Billion molecules against COVID‐19 challenge”, to identify small‐molecule inhibitors against SARS‐CoV‐2 or relevant human receptors. Participating teams used a wide variety of computational methods to screen a minimum of 1 billion virtual molecules against 6 protein targets. Overall, 31 teams participated, and they suggested a total of 639,024 molecules, which were subsequently ranked to find ‘consensus compounds’. The organizing team coordinated with various contract research organizations (CROs) and collaborating institutions to synthesize and test 878 compounds for biological activity against proteases (Nsp5, Nsp3, TMPRSS2), nucleocapsid N, RdRP (only the Nsp12 domain), and (alpha) spike protein S. Overall, 27 compounds with weak inhibition/binding were experimentally identified by binding‐, cleavage‐, and/or viral suppression assays and are presented here. Open science approaches such as the one presented here contribute to the knowledge base of future drug discovery efforts in finding better SARS‐CoV‐2 treatments.

Equivariant neural networks, whose hidden features transform according to representations of a group G acting on the data, exhibit training efficiency and an improved generalisation performance. In this work, we extend group invariant and equivariant representation learning to the field of unsupervised deep learning. We propose a general learning strategy based on an encoder-decoder framework in which the latent representation is separated in an invariant term and an equivariant group action component. The key idea is that the network learns to encode and decode data to and from a group-invariant representation by additionally learning to predict the appropriate group action to align input and output pose to solve the reconstruction task. We derive the necessary conditions on the equivariant encoder, and we present a construction valid for any G, both discrete and continuous. We describe explicitly our construction for rotations, translations and permutations. We test the validity and the robustness of our approach in a variety of experiments with diverse data types employing different network architectures.

Recently, there has been great success in applying deep neural networks on graph structured data. Most work, however, focuses on either nodeor graph-level supervised learning, such as node, link or graph classification or node-level unsupervised learning (e.g. node clustering). Despite its wide range of possible applications, graphlevel unsupervised learning has not received much attention yet. This might be mainly attributed to the high representation complexity of graphs, which can be represented by n! equivalent adjacency matrices, where n is the number of nodes. In this work we address this issue by proposing a permutation-invariant variational autoencoder for graph structured data. Our proposed model indirectly learns to match the node ordering of input and output graph, without imposing a particular node ordering or performing expensive graph matching. We demonstrate the effectiveness of our proposed model on various graph reconstruction and generation tasks and evaluate the expressive power of extracted representations for downstream graph-level classification and regression.

In this work we introduce an Autoencoder for molecular conformations. Our proposed model converts the discrete spatial arrangements of atoms in a given molecular graph (conformation) into and from a continuous fixed-sized latent representation. We demonstrate that in this latent representation, similar conformations cluster together while distinct conformations split apart. Moreover, by training a probabilistic model on a large dataset of molecular conformations, we demonstrate how our model can be used to generate diverse sets of energetically favorable conformations for a given molecule. Finally, we show that the continuous representation allows us to utilize optimization methods to find molecules that have conformations with favourable spatial properties.

Automatic recognition of the molecular content of a molecule’s graphical depiction is an extremely challenging problem that remains largely unsolved despite decades of research. Recent advances in neural machine translation enable the auto-encoding of molecular structures in a continuous vector space of fixed size (latent representation) with low reconstruction errors. In this paper, we present a fast and accurate model combining a deep convolutional neural network learning from molecule depictions and a pre-trained decoder that translates the latent representation into the SMILES representation of the molecules. This combination allows to precisely infer a molecular structure from an image. Our rigorous evaluation show that Img2Mol is able to correctly translate up to 88% of the molecular depictions into their SMILES representation. A pretrained version of Img2Mol is made publicly available on GitHub for non-commercial users.

In silico protein–ligand binding prediction is an ongoing area of research in computational chemistry and machine learning based drug discovery, as an accurate predictive model could greatly reduce the time and resources necessary for the detection and prioritization of possible drug candidates. Proteochemometric modeling (PCM) attempts to create an accurate model of the protein–ligand interaction space by combining explicit protein and ligand descriptors. This requires the creation of information-rich, uniform and computer interpretable representations of proteins and ligands. Previous studies in PCM modeling rely on pre-defined, handcrafted feature extraction methods, and many methods use protein descriptors that require alignment or are otherwise specific to a particular group of related proteins. However, recent advances in representation learning have shown that unsupervised machine learning can be used to generate embeddings that outperform complex, human-engineered representations. Several different embedding methods for proteins and molecules have been developed based on various language-modeling methods. Here, we demonstrate the utility of these unsupervised representations and compare three protein embeddings and two compound embeddings in a fair manner. We evaluate performance on various splits of a benchmark dataset, as well as on an internal dataset of protein–ligand binding activities and find that unsupervised-learned representations significantly outperform handcrafted representations.

Protecting molecular structures from disclosure against external parties is of great relevance for industrial and private associations, such as pharmaceutical companies. Within the framework of external collaborations, it is common to exchange datasets by encoding the molecular structures into descriptors. Molecular fingerprints such as the extended-connectivity fingerprints are frequently used for such an exchange, because they typically perform well on quantitative structure-activity relationship tasks. ECFPs are often considered to be non-invertible due to the way they are computed.In this paper, we present a reverse-engineering method to deduce the molecular structure given revealed ECFPs. Our method includes the Neuraldecipher , a neural network model that predicts a compact vector representation of compounds, given ECFPs. We then utilize another pre-trained model to retrieve the molecular structure as SMILES representation. We demonstrate that our method is able to reconstruct molecular structures to some extent, and improves, when ECFPs with larger fingerprint sizes are revealed. For example, given ECFP count vectors of length 4096, we are able to correctly deduce around 60% of molecular structures on a validation set (112K unique samples) with our method.

Protecting molecular structures from disclosure against external parties is of great relevance for industrial and private associations, such as pharmaceutical companies. Within the framework of external collaborations, it is common to exchange datasets by encoding the molecular structures into descriptors. Molecular fingerprints such as the extended-connectivity fingerprints (ECFPs) are frequently used for such an exchange, because they typically perform well on quantitative structure-activity relationship tasks. ECFPs are often considered to be non-invertible due to the way they are computed. In this paper, we present a fast reverse-engineering method to deduce the molecular structure given revealed ECFPs. Our method includes the Neuraldecipher, a neural network model that predicts a compact vector representation of compounds, given ECFPs. We then utilize another pre-trained model to retrieve the molecular structure as SMILES representation. We demonstrate that our method is able to reconstruct molecular structures to some extent, and improves, when ECFPs with larger fingerprint sizes are revealed. For example, given ECFP count vectors of length 4096, we are able to correctly deduce up to 69% of molecular structures on a validation set (112 K unique samples) with our method.

In-silico protein-ligand binding prediction is an ongoing area of research in computational chemistry and machine learning based drug discovery, as an accurate predictive model could greatly reduce the time and resources necessary for the detection and prioritization of possible drug candidates. Proteochemometric modeling (PCM) attempts to make an accurate model of the protein-ligand interaction space by combining explicit protein and ligand descriptors. This requires the creation of information-rich, uniform and computer interpretable representations of proteins and ligands. Previous work in PCM modeling relies on pre-defined, handcrafted feature extraction methods, and many methods use protein descriptors that require alignment or are otherwise specific to a particular group of related proteins. However, recent advances in representation learning have shown that unsupervised machine learning can be used to generate embeddings which outperform complex, human-engineered representations. We apply this reasoning to propose a novel proteochemometric modeling methodology which, for the first time, uses embeddings generated via unsupervised representation learning for both the protein and ligand descriptors. We evaluate performance on various splits of a benchmark dataset, including a challenging split that tests the model’s ability to generalize to proteins for which bioactivity data is greatly limited, and we find that our method consistently outperforms state-of-the-art methods.

Summary Optimizing small molecules in a drug discovery project is a notoriously difficult task as multiple molecular properties have to be considered and balanced at the same time. In this work, we present our novel interactive in silico compound optimization platform termed grünifai to support the ideation of the next generation of compounds under the constraints of a multiparameter objective. grünifai integrates adjustable in silico models, a continuous representation of the chemical space, a scalable particle swarm optimization algorithm and the possibility to actively steer the compound optimization through providing feedback on generated intermediate structures. Availability and implementation Source code and documentation are freely available under an MIT license and are openly available on GitHub (https://github.com/jrwnter/gruenifai). The backend, including the optimization method and distribution on multiple GPU nodes is written in Python 3. The frontend is written in ReactJS.

We utilize Particle Swarm Optimization to optimize molecules in a machine-learned continuous chemical representation with respect to multiple objectives such as biological activity, structural constrains or ADMET properties.

Generative adversarial networks (GANs) are a powerful framework for generative tasks. However, they are difficult to train and tend to miss modes of the true data generation process. Although GANs can learn a rich representation of the covered modes of the data in their latent space, the framework misses an inverse mapping from data to this latent space. We propose Invariant Encoding Generative Adversarial Networks (IVE-GANs), a novel GAN framework that introduces such a mapping for individual samples from the data by utilizing features in the data which are invariant to certain transformations. Since the model maps individual samples to the latent space, it naturally encourages the generator to cover all modes. We demonstrate the effectiveness of our approach in terms of generative performance and learning rich representations on several datasets including common benchmark image generation tasks.

There has been a recent surge of interest in using machine learning across chemical space in order to predict properties of molecules or design molecules and materials with desired properties. Most of this work relies on defining clever feature representations, in which the chemical graph structure is encoded in a uniform way such that predictions across chemical space can be made. In this work, we propose to exploit the powerful ability of deep neural networks to learn a feature representation from low-level encodings of a huge corpus of chemical structures. Our model borrows ideas from neural machine translation: it translates between two semantically equivalent but syntactically different representations of molecular structures, compressing the meaningful information both representations have in common in a low-dimensional representation vector. Once the model is trained, this representation can be extracted for any new molecule and utilized as descriptor. In fair benchmarks with respect to various human-engineered molecular fingerprints and graph-convolution models, our method shows competitive performance in modelling quantitative structure-activity relationships in all analyzed datasets. Additionally, we show that our descriptor significantly outperforms all baseline molecular fingerprints in two ligand-based virtual screening tasks. Overall, our descriptors show the most consistent performances over all experiments. The continuity of the descriptor space and the existence of the decoder that permits to deduce a chemical structure from an embedding vector allows for exploration of the space and opens up new opportunities for compound optimization and idea generation.