Olexandr Isayev

Materials Cartography: Representing and Mining Materials Space Using Structural and Electronic Fingerprints

As the proliferation of high-throughput approaches in materials science is increasing the wealth of data in the field, the gap between accumulated-information and derived-knowledge widens. We address the issue of scientific discovery in materials databases by introducing novel analytical approaches based on structural and electronic materials fingerprints. The framework is employed to (i) query large databases of materials using similarity concepts, (ii) map the connectivity of materials space (i.e., as a materials cartograms) for rapidly identifying regions with unique organizations/properties, and (iii) develop predictive Quantitative Materials Structure–Property Relationship models for guiding materials design. In this study, we test these fingerprints by seeking target material properties. As a quantitative example, we model the critical temperatures of known superconductors. Our novel materials fingerprinting and materials cartography approaches contribute to the emerging field of materials informati...

Universal fragment descriptors for predicting properties of inorganic crystals

Although historically materials discovery has been driven by a laborious trial-and-error process, knowledge-driven materials design can now be enabled by the rational combination of Machine Learning methods and materials databases. Here, data from the AFLOW repository for ab initio calculations is combined with Quantitative Materials Structure-Property Relationship models to predict important properties: metal/insulator classification, band gap energy, bulk/shear moduli, Debye temperature and heat capacities. The predictions accuracy compares well with the quality of the training data for virtually any stoichiometric inorganic crystalline material, reciprocating the available thermomechanical experimental data. The universality of the approach is attributed to the construction of the descriptors: Property-Labelled Materials Fragments. The representations require only minimal structural input allowing straightforward implementations of simple heuristic design rules.

QSAR Modeling of Tox21 Challenge Stress Response and Nuclear Receptor Signaling Toxicity Assays

The ability to determine which environmental chemicals pose the greatest potential threats to human health remains one of the major concerns in regulatory toxicology. Computation methods that can accurately predict the chemicals’ toxic potential in silico are increasingly sought-after to replace in vitro high-throughput screening (HTS) as well as controversial and costly in vivo animal studies. To this end, we have built Quantitative Structure-Activity Relationship (QSAR) models of twelve (12) stress response and nuclear receptor signaling pathways toxicity assays as part of the 2014 Tox21 Challenge. Our models were built using the Random Forest, Deep Neural Networks and various combinations of descriptors and balancing protocols. All of our models were statistically significant for each of the 12 assays with the balanced accuracy in the range between 0.58 and 0.82. Our results also show that models built with Deep Neural Networks had high accuracy than those developed with simple machine learning algorithms and that dataset balancing led to a significant accuracy decrease.

ANI-1, A data set of 20 million calculated off-equilibrium conformations for organic molecules

One of the grand challenges in modern theoretical chemistry is designing and implementing approximations that expedite ab initio methods without loss of accuracy. Machine learning (ML) methods are emerging as a powerful approach to constructing various forms of transferable atomistic potentials. They have been successfully applied in a variety of applications in chemistry, biology, catalysis, and solid-state physics. However, these models are heavily dependent on the quality and quantity of data used in their fitting. Fitting highly flexible ML potentials, such as neural networks, comes at a cost: a vast amount of reference data is required to properly train these models. We address this need by providing access to a large computational DFT database, which consists of more than 20 M off equilibrium conformations for 57,462 small organic molecules. We believe it will become a new standard benchmark for comparison of current and future methods in the ML potential community.

Are the reduction and oxidation properties of nitrocompounds dissolved in water different from those produced when adsorbed on a silica surface? A DFT M05-2X computational study

The reduction and oxidation properties of four nitrocompounds (trinitrotoluene [TNT], 2,4‐dinitrotoluene, 2,4‐dinitroanisole, and 5‐nitro‐2,4‐dihydro‐3H‐1,2,4‐triazol‐3‐one [NTO]) dissolved in water as compared with the same properties for compounds adsorbed on a silica surface were studied. To consider the influence of adsorption, cluster models were developed at the M05/tzvp level. A hydroxylated silica (001) surface was chosen to represent a key component of soil. The PCM(Pauling) and SMD solvation models were used to model water bulk influence. The following properties were analyzed: electron affinity, ionization potential, reduction Gibbs free energy, oxidation Gibbs free energy, and reduction and oxidation potentials. It was found that adsorption and solvation decrease gas phase electron affinity, ionization potential, and Gibbs free energy of reduction and oxidation, and thus, promote redox transformation of nitrocompounds. However, in case of solvation, the changes are more significant than for adsorption. This means that nitrocompounds dissolved in water are easier to transform by reduction or oxidation than adsorbed ones. Among the considered compounds, TNT was found to be the most reactive in an electron attachment process and the least reactive for an electron detachment transformation. During ionization, a deprotonation of adsorbed NTO was found to occur.

Deep reinforcement learning for de novo drug design

We introduce an artificial intelligence approach to de novo design of molecules with desired physical or biological properties. We have devised and implemented a novel computational strategy for de novo design of molecules with desired properties termed ReLeaSE (Reinforcement Learning for Structural Evolution). On the basis of deep and reinforcement learning (RL) approaches, ReLeaSE integrates two deep neural networks—generative and predictive—that are trained separately but are used jointly to generate novel targeted chemical libraries. ReLeaSE uses simple representation of molecules by their simplified molecular-input line-entry system (SMILES) strings only. Generative models are trained with a stack-augmented memory network to produce chemically feasible SMILES strings, and predictive models are derived to forecast the desired properties of the de novo–generated compounds. In the first phase of the method, generative and predictive models are trained separately with a supervised learning algorithm. In the second phase, both models are trained jointly with the RL approach to bias the generation of new chemical structures toward those with the desired physical and/or biological properties. In the proof-of-concept study, we have used the ReLeaSE method to design chemical libraries with a bias toward structural complexity or toward compounds with maximal, minimal, or specific range of physical properties, such as melting point or hydrophobicity, or toward compounds with inhibitory activity against Janus protein kinase 2. The approach proposed herein can find a general use for generating targeted chemical libraries of novel compounds optimized for either a single desired property or multiple properties.

Machine learning for molecular and materials science

Here we summarize recent progress in machine learning for the chemical sciences. We outline machine-learning techniques that are suitable for addressing research questions in this domain, as well as future directions for the field. We envisage a future in which the design, synthesis, characterization and application of molecules and materials is accelerated by artificial intelligence.Recent progress in machine learning in the chemical sciences and future directions in this field are discussed.

Less is more: Sampling chemical space with active learning

The development of accurate and transferable machine learning (ML) potentials for predicting molecular energetics is a challenging task. The process of data generation to train such ML potentials is a task neither well understood nor researched in detail. In this work, we present a fully automated approach for the generation of datasets with the intent of training universal ML potentials. It is based on the concept of active learning (AL) via Query by Committee (QBC), which uses the disagreement between an ensemble of ML potentials to infer the reliability of the ensembles prediction. QBC allows the presented AL algorithm to automatically sample regions of chemical space where the ML potential fails to accurately predict the potential energy. AL improves the overall fitness of ANAKIN-ME (ANI) deep learning potentials in rigorous test cases by mitigating human biases in deciding what new training data to use. AL also reduces the training set size to a fraction of the data required when using naive random sampling techniques. To provide validation of our AL approach, we develop the COmprehensive Machine-learning Potential (COMP6) benchmark (publicly available on GitHub) which contains a diverse set of organic molecules. Active learning-based ANI potentials outperform the original random sampled ANI-1 potential with only 10% of the data, while the final active learning-based model vastly outperforms ANI-1 on the COMP6 benchmark after training to only 25% of the data. Finally, we show that our proposed AL technique develops a universal ANI potential (ANI-1x) that provides accurate energy and force predictions on the entire COMP6 benchmark. This universal ML potential achieves a level of accuracy on par with the best ML potentials for single molecules or materials, while remaining applicable to the general class of organic molecules composed of the elements CHNO.

Discovering a Transferable Charge Assignment Model Using Machine Learning

Partial atomic charge assignment is of immense practical value to force field parametrization, molecular docking, and cheminformatics. Machine learning has emerged as a powerful tool for modeling chemistry at unprecedented computational speeds given accurate reference data. However, certain tasks, such as charge assignment, do not have a unique solution. Herein, we use a machine learning algorithm to discover a new charge assignment model by learning to replicate molecular dipole moments across a large, diverse set of nonequilibrium conformations of molecules containing C, H, N, and O atoms. The new model, called Affordable Charge Assignment (ACA), is computationally inexpensive and predicts dipoles of out-of-sample molecules accurately. Furthermore, dipole-inferred ACA charges are transferable to dipole and even quadrupole moments of much larger molecules than those used for training. We apply ACA to dynamical trajectories of biomolecules and produce their infrared spectra. Additionally, we find that ACA assigns similar charges to Charge Model 5 but with greatly reduced computational cost.

Transferable Dynamic Molecular Charge Assignment Using Deep Neural Networks

The ability to accurately and efficiently compute quantum-mechanical partial atomistic charges has many practical applications, such as calculations of IR spectra, analysis of chemical bonding, and classical force field parametrization. Machine learning (ML) techniques provide a possible avenue for the efficient prediction of atomic partial charges. Modern ML advances in the prediction of molecular energies [i.e., the hierarchical interacting particle neural network (HIP-NN)] has provided the necessary model framework and architecture to predict transferable, extensible, and conformationally dynamic atomic partial charges based on reference density functional theory (DFT) simulations. Utilizing HIP-NN, we show that ML charge prediction can be highly accurate over a wide range of molecules (both small and large) across a variety of charge partitioning schemes such as the Hirshfeld, CM5, MSK, and NBO methods. To demonstrate transferability and size extensibility, we compare ML results with reference DFT calculations on the COMP6 benchmark, achieving errors of 0.004e- (elementary charge). This is remarkable since this benchmark contains two proteins that are multiple times larger than the largest molecules in the training set. An application of our atomic charge predictions on nonequilibrium geometries is the generation of IR spectra for organic molecules from dynamical trajectories on a variety of organic molecules, which show good agreement with calculated IR spectra with reference method. Critically, HIP-NN charge predictions are many orders of magnitude faster than direct DFT calculations. These combined results provide further evidence that ML (specifically HIP-NN) provides a pathway to greatly increase the range of feasible simulations while retaining quantum-level accuracy.