Molecular Generative Model Based on an Adversarially Regularized Autoencoder.

Deep generative models are attracting great attention as a new promising approach for molecular design. A variety of models reported so far are based on either a variational autoencoder (VAE) or a generative adversarial network (GAN), but they have limitations such as low validity and uniqueness. Here, we propose a new type of model based on an adversarially regularized autoencoder (ARAE). It basically uses latent variables like VAE, but the distribution of the latent variables is estimated by adversarial training like in GAN. The latter is intended to avoid both the insufficiently flexible approximation of posterior distribution in VAE and the difficulty in handling discrete variables in GAN. Our benchmark study showed that ARAE indeed outperformed conventional models in terms of validity, uniqueness, and novelty per generated molecule. We also demonstrated a successful conditional generation of drug-like molecules with ARAE for the control of both cases of single and multiple properties. As a potential real-world application, we could generate epidermal growth factor receptor inhibitors sharing the scaffolds of known active molecules while satisfying drug-like conditions simultaneously.

Comprehensive Study on Molecular Supervised Learning with Graph Neural Networks.

This work considers strategies to develop accurate and reliable graph neural networks (GNNs) for molecular property predictions. Prediction performance of GNNs is highly sensitive to the change in various parameters due to the inherent challenges in molecular machine learning, such as a deficient amount of data samples and bias in data distribution. Comparative studies with well-designed experiments are thus important to clearly understand which GNNs are powerful for molecular supervised learning. Our work presents a number of ablation studies along with a guideline to train and utilize GNNs for both molecular regression and classification tasks. First, we validate that using both atomic and bond meta-information improves the prediction performance in the regression task. Second, we find that the graph isomorphism hypothesis proposed by [Xu, K.; et al How powerful are graph neural networks? 2018, arXiv:1810.00826. arXiv.org e-Print archive. https://arxiv.org/abs/1810.00826] is valid for the regression task. Surprisingly, however, the findings above do not hold for the classification tasks. Beyond the study on model architectures, we test various regularization methods and Bayesian learning algorithms to find the best strategy to achieve a reliable classification system. We demonstrate that regularization methods penalizing predictive entropy might not give well-calibrated probability estimation, even though they work well in other domains, and Bayesian learning methods are capable of developing reliable prediction systems. Furthermore, we argue the importance of Bayesian learning in virtual screening by showing that well-calibrated probability estimation may lead to a higher success rate.

Predicting Drug-Target Interaction Using a Novel Graph Neural Network with 3D Structure-Embedded Graph Representation.

We propose a novel deep learning approach for predicting drug-target interaction using a graph neural network. We introduce a distance-aware graph attention algorithm to differentiate various types of intermolecular interactions. Furthermore, we extract the graph feature of intermolecular interactions directly from the 3D structural information on the protein-ligand binding pose. Thus, the model can learn key features for accurate predictions of drug-target interaction rather than just memorize certain patterns of ligand molecules. As a result, our model shows better performance than docking and other deep learning methods for both virtual screening (AUROC of 0.968 for the DUD-E test set) and pose prediction (AUROC of 0.935 for the PDBbind test set). In addition, it can reproduce the natural population distribution of active molecules and inactive molecules.

Effects of the locality of a potential derived from hybrid density functionals on Kohn-Sham orbitals and excited states.

Density functional theory (DFT) has been an essential tool for electronic structure calculations in various fields. In particular, its hybrid method including the Hartree-Fock (HF) exchange term remarkably improves the reliability of DFT for chemical applications and computational material design. There are two different types of exchange-correlation potential that can be derived from hybrid functionals. The conventional approach adopts a non-multiplicative potential including the non-local HF exchange operator. Herein, we propose to use a local multiplicative potential as an alternative for accurate excited state calculations. We show that such a local potential can be derived from existing global hybrid functionals using the optimized effective potential method. As a proof-of-concept, we chose PBE0 and investigated its performance for the Caricato benchmark set. Unlike the conventional one, the local potential produced orbital energy gaps with no strong dependence on the mixing ratio as a good approximation for optical excitations. Furthermore, its time-dependent DFT resulted in a surprisingly small mean absolute error even with a local density approximation kernel, surpassing all reported values with various popular functionals. In particular, most excitations were dictated by single orbital transitions due to physically meaningful virtual orbitals, which is beneficial to clear interpretations in the molecular orbital picture.

Supersampling method for efficient grid-based electronic structure calculations.

The egg-box effect, the spurious variation of energy and force due to the discretization of continuous space, is an inherent vexing problem in grid-based electronic structure calculations. Its effective suppression allowing for large grid spacing is thus crucial for accurate and efficient computations. We here report that the supersampling method drastically alleviates it by eliminating the rapidly varying part of a target function along both radial and angular directions. In particular, the use of the sinc filtering function performs best because as an ideal low pass filter it clearly cuts out the high frequency region beyond allowed by a given grid spacing.

GalaxyDock-DL: Protein-Ligand Docking by Global Optimization and Neural Network Energy.

With the recent introduction of deep learning techniques into the prediction of biomolecular structures, structure prediction performance has significantly improved, and the potential for biomedical applications has increased considerably. The prediction of protein-ligand complex structures, applicable to the atomistic understanding of biomolecular functions and the effective design of drug molecules, has also improved with the introduction of deep learning. In this paper, it is demonstrated that docking performance can be greatly enhanced by training an energy function that encapsulates physical effects using deep learning within the framework of the traditional protein-ligand docking method. The advantage of this method, called GalaxyDock-DL, lies in its minimal overfitting to the training data compared to several existing deep learning-based protein-ligand docking methods. Unlike some recent deep learning methods, it does not use information about known binding pocket center positions. Instead, the results of this docking method show a systematic dependence on the physical properties of the target protein-ligand complexes such as atomic thermal fluctuations and binding affinity. GalaxyDock-DL utilizes the global optimization technique of the conventional protein-ligand docking method, GalaxyDock, and a neural network energy function trained to stabilize the native state compared to non-native states, just as physical free energy does. This physical principle-based approach suggests directions not only for future structure prediction involving structurally flexible biomolecular complexes but also for predicting binding affinity, thereby providing guidance for the effective design of biofunctional ligands.

A Bayesian graph convolutional network for reliable prediction of molecular properties with uncertainty quantification.

Deep neural networks have been increasingly used in various chemical fields. In the nature of a data-driven approach, their performance strongly depends on data used in training. Therefore, models developed in data-deficient situations can cause highly uncertain predictions, leading to vulnerable decision making. Here, we show that Bayesian inference enables more reliable prediction with quantitative uncertainty analysis. Decomposition of the predictive uncertainty into model- and data-driven uncertainties allows us to elucidate the source of errors for further improvements. For molecular applications, we devised a Bayesian graph convolutional network (GCN) and evaluated its performance for molecular property predictions. Our study on the classification problem of bio-activity and toxicity shows that the confidence of prediction can be quantified in terms of the predictive uncertainty, leading to more accurate virtual screening of drug candidates than standard GCNs. The result of log P prediction illustrates that data noise affects the data-driven uncertainty more significantly than the model-driven one. Based on this finding, we could identify artefacts that arose from quantum mechanical calculations in the Harvard Clean Energy Project dataset. Consequently, the Bayesian GCN is critical for molecular applications under data-deficient conditions.

Molecular generative model based on conditional variational autoencoder for de novo molecular design.

We propose a molecular generative model based on the conditional variational autoencoder for de novo molecular design. It is specialized to control multiple molecular properties simultaneously by imposing them on a latent space. As a proof of concept, we demonstrate that it can be used to generate drug-like molecules with five target properties. We were also able to adjust a single property without changing the others and to manipulate it beyond the range of the dataset.