Stochastic machine learning via sigma profiles to build a digital chemical space.

This work establishes a different paradigm on digital molecular spaces and their efficient navigation by exploiting sigma profiles. To do so, the remarkable capability of Gaussian processes (GPs), a type of stochastic machine learning model, to correlate and predict physicochemical properties from sigma profiles is demonstrated, outperforming state-of-the-art neural networks previously published. The amount of chemical information encoded in sigma profiles eases the learning burden of machine learning models, permitting the training of GPs on small datasets which, due to their negligible computational cost and ease of implementation, are ideal models to be combined with optimization tools such as gradient search or Bayesian optimization (BO). Gradient search is used to efficiently navigate the sigma profile digital space, quickly converging to local extrema of target physicochemical properties. While this requires the availability of pretrained GP models on existing datasets, such limitations are eliminated with the implementation of BO, which can find global extrema with a limited number of iterations. A remarkable example of this is that of BO toward boiling temperature optimization. Holding no knowledge of chemistry except for the sigma profile and boiling temperature of carbon monoxide (the worst possible initial guess), BO finds the global maximum of the available boiling temperature dataset (over 1,000 molecules encompassing more than 40 families of organic and inorganic compounds) in just 15 iterations (i.e., 15 property measurements), cementing sigma profiles as a powerful digital chemical space for molecular optimization and discovery, particularly when little to no experimental data is initially available.

Consistent and reproducible computation of the glass transition temperature from molecular dynamics simulations.

In many fields, from semiconductors for opto-electronic applications to ionic liquids (ILs) for separations, the glass transition temperature (Tg) of a material is a useful gauge for its potential use in practical settings. As a result, there is a great deal of interest in predicting Tg using molecular simulations. However, the uncertainty and variation in the trend shift method, a common approach in simulations to predict Tg, can be high. This is due to the need for human intervention in defining a fitting range for linear fits of density with temperature assumed for the liquid and glass phases across the simulated cooling. The definition of such fitting ranges then defines the estimate for the Tg as the intersection of linear fits. We eliminate this need for human intervention by leveraging the Shapiro-Wilk normality test and proposing an algorithm to define the fitting ranges and, consequently, Tg. Through this integration, we incorporate into our automated methodology that residuals must be normally distributed around zero for any fit, a requirement that must be met for any regression problem. Consequently, fitting ranges for realizing linear fits for each phase are statistically defined rather than visually inferred, obtaining an estimate for Tg without any human intervention. The method is also capable of finding multiple linear regimes across density vs temperature curves. We compare the predictions of our proposed method across multiple IL and semiconductor molecular dynamics simulation results from the literature and compare other proposed methods for automatically detecting Tg from density-temperature data. We believe that our proposed method would allow for more consistent predictions of Tg. We make this methodology available and open source through GitHub.

Machine Learning Identification of Organic Compounds Using Visible Light.

Identifying chemical compounds is essential in several areas of science and engineering. Laser-based techniques are promising for autonomous compound detection because the optical response of materials encodes enough electronic and vibrational information for remote chemical identification. This has been exploited using the fingerprint region of infrared absorption spectra, which involves a dense set of absorption peaks that are unique to individual molecules, thus facilitating chemical identification. However, optical identification using visible light has not been realized. Using decades of experimental refractive index data in the scientific literature of pure organic compounds and polymers over a broad range of frequencies from the ultraviolet to the far-infrared, we develop a machine learning classifier that can accurately identify organic species based on a single-wavelength dispersive measurement in the visible spectral region, away from absorption resonances. The optical classifier proposed here could be applied to autonomous material identification protocols and applications.

Atomistic simulation of soft porous coordination polymers.

Soft porous coordination polymers (SPCPs) are materials with exceptional potential because of their ability to incorporate the properties of nominally rigid porous materials like metal-organic frameworks (MOFs) and those of soft matter, such as polymers of intrinsic microporosity (PIMs). This combination could offer the gas adsorption properties of MOFs together with the mechanical stability and processability of PIMs, opening up a space of flexible, highly responsive adsorbing materials. In order to understand their structure and behavior, we present a process for the construction of amorphous SPCPs from secondary building blocks. We then use classical molecular dynamics simulations to characterize the resulting structures based on branch functionalities (f), pore size distributions (PSDs), and radial distribution functions and compare them to experimentally synthesized analogs. In the course of this comparison, we demonstrate that the pore structure of SPCPs is due to both pores intrinsic to the secondary building blocks, and intercolloid spacing between colloid particles. We also illustrate the differences in nanoscale structure based on linker length and flexibility, particularly in the PSDs, finding that stiff linkers tend to produce SPCPs with larger maximum pore sizes.

The effect of co-adsorbed solvent molecules on H2 binding to metal alkoxides.

The introduction of metal alkoxides has been proposed as an attractive option to enhance hydrogen binding energies in porous materials such as metal-organic frameworks (MOFs) for room-temperature hydrogen storage applications. The presence of residual solvent molecules from MOF synthesis can, however, affect the performance of these functional groups. We perform quantum chemical calculations to predict solvent binding energies onto the metal-alkoxides and the temperatures required to drive off the solvent molecules and successfully activate porous materials with these moieties. Calculations are performed for Li, Mg, Zn, Cu, and Ni alkoxides and chloroform (CHCl3), dimethylformamide (DMF), ethanol, methanol, and water solvent molecules. We identify CHCl3 as a promising solvent that can be removed from these alkoxides at mild temperatures, whereas DMF binds strongly to the metal alkoxides and removal would require temperatures above the present upper bound of thermal stability in MOFs. As a second objective, we calculated the binding energies of hydrogen to metal alkoxide-solvent complexes to explore the effect of any solvent molecules that cannot be removed.

Free energy of metal-organic framework self-assembly.

Metal-organic frameworks (MOFs) represent an important class of materials. Careful selection of building blocks allows for tailoring of the properties of the resulting framework. The self-assembly process, however, is not understood, and without detailed knowledge of the underlying molecular mechanism, it is difficult to anticipate whether a particular design can be realized, or whether the material adopts a metastable, kinetically arrested state. We present a detailed examination of early-stage self-assembly pathways of the MOF-5. Enhanced sampling techniques are used to model a self-assembly in an explicit solvent (dimethylformamide, DMF). We identify several free energy barriers encountered during the assembly of the final MOF, which arise from structural rearrangements preceding MOF formation and from disrupted MOF-solvent interactions as formation proceeds. In all cases considered here, MOFs exhibit favorable entropic gains during the assembly. More generally, the strategy presented provides a step toward the experimental design characterizing the formation of ordered frameworks and possible sources of polymorphism.

SSAGES: Software Suite for Advanced General Ensemble Simulations.

Molecular simulation has emerged as an essential tool for modern-day research, but obtaining proper results and making reliable conclusions from simulations requires adequate sampling of the system under consideration. To this end, a variety of methods exist in the literature that can enhance sampling considerably, and increasingly sophisticated, effective algorithms continue to be developed at a rapid pace. Implementation of these techniques, however, can be challenging for experts and non-experts alike. There is a clear need for software that provides rapid, reliable, and easy access to a wide range of advanced sampling methods and that facilitates implementation of new techniques as they emerge. Here we present SSAGES, a publicly available Software Suite for Advanced General Ensemble Simulations designed to interface with multiple widely used molecular dynamics simulations packages. SSAGES allows facile application of a variety of enhanced sampling techniques-including adaptive biasing force, string methods, and forward flux sampling-that extract meaningful free energy and transition path data from all-atom and coarse-grained simulations. A noteworthy feature of SSAGES is a user-friendly framework that facilitates further development and implementation of new methods and collective variables. In this work, the use of SSAGES is illustrated in the context of simple representative applications involving distinct methods and different collective variables that are available in the current release of the suite. The code may be found at: https://github.com/MICCoM/SSAGES-public.

High-throughput computational screening of metal-organic frameworks.

There is an almost unlimited number of metal-organic frameworks (MOFs). This creates exciting opportunities but also poses a problem: how do we quickly find the best MOFs for a given application? Molecular simulations have advanced sufficiently that many MOF properties - especially structural and gas adsorption properties - can be predicted computationally, and molecular modeling techniques are now used increasingly to guide the synthesis of new MOFs. With increasing computational power and improved simulation algorithms, it has become possible to conduct high-throughput computational screening to identify promising MOF structures and uncover structure-property relations. We review these efforts and discuss future directions in this new field.

Combining High-Throughput Experiments and Active Learning to Characterize Deep Eutectic Solvents.

The high tunability of deep eutectic solvents (DESs) stems from the ease of changing their precursors and relative compositions. However, measuring the physicochemical properties across large composition and temperature ranges, necessary to properly design target-specific DESs, is tedious and error-prone and represents a bottleneck in the advancement and scalability of DES-based applications. As such, active learning (AL) methodologies based on Gaussian processes (GPs) were developed in this work to minimize the experimental effort necessary to characterize DESs. Owing to its importance for large-scale applications, the reduction of DES viscosity through the addition of a low-molecular-weight solvent was explored as a case study. A high-throughput experimental screening was initially performed on nine different ternary DESs. Then, GPs were successfully trained to predict DES viscosity from its composition and temperature, showcasing the ability of these stochastic, nonparametric models to accurately describe the physicochemical properties of complex mixtures. Finally, the ability of GPs to provide estimates of their own uncertainty was leveraged through an AL framework to minimize the number of data points necessary to obtain accurate viscosity modes. This led to a significant reduction in data requirements, with many systems requiring only five independent viscosity data points to be properly described.

Boosting Graph Neural Networks with Molecular Mechanics: A Case Study of Sigma Profile Prediction.

Sigma profiles are quantum-chemistry-derived molecular descriptors that encode the polarity of molecules. They have shown great performance when used as a feature in machine learning applications. To accelerate the development of these models and the construction of large sigma profile databases, this work proposes a graph convolutional network (GCN) architecture to predict sigma profiles from molecule structures. To do so, the usage of molecular mechanics (force field atom types) is explored as a computationally inexpensive node-level featurization technique to encode the local and global chemical environments of atoms in molecules. The GCN models developed in this work accurately predict the sigma profiles of assorted organic and inorganic compounds. The best GCN model here reported, obtained using Merck molecular force field (MMFF) atom types, displayed training and testing set coefficients of determination of 0.98 and 0.96, respectively, which are superior to previous methodologies reported in the literature. This performance boost is shown to be due to both the usage of a convolutional architecture and node-level features based on force field atom types. Finally, to demonstrate their practical applicability, we used GCN-predicted sigma profiles as the input to machine learning models previously developed in the literature that predict boiling temperatures and aqueous solubilities. Using the predicted sigma profiles as input, these models were able to compute both physicochemical properties using significantly less computational resources and displayed only a slight decrease in performance when compared with sigma profiles obtained from quantum chemistry methods.