Computer Modeling and Machine Learning in Chemistry and Materials Science: From Properties and Reactions of Small Organic and Inorganic Molecules to the Smart Design of Polymers and Composites ^†

Abstract

Computer modeling, machine learning, and artificial intelligence are currently considered cutting-edge topics in chemistry and materials science. The application of information technologies in natural sciences can help researchers collect big data and understand patterns that are not obvious to humans. In this perspective, I would like to highlight the recent achievements of our research group and other researchers in relation to computer modeling and machine learning in chemistry and materials science.

Computer modeling, machine learning, and artificial intelligence are cutting-edge topics in chemistry and materials science today (Figure 1). The application of information technologies in natural sciences can help researchers collect big data and understand patterns that are not obvious to humans. Interdisciplinary studies in the fields of computational and quantum chemistry, computational statistics, artificial intelligence, machine learning, neural networks, predictive analytics, data mining, and data science are pivotal to the future of modern theoretical chemistry. Understanding the relationship between the structures, properties, or functions of all kinds of compounds (including mechanical, thermal, structural, electric, magnetic, and optical properties), as well as development of various theories, is the paradigm of natural sciences (and, in particular, chemistry).

My main field of expertise, computational chemistry, is a very powerful tool that can be used for studying the structures, properties, and reactivity levels of biologically active and small organic/inorganic molecules. In the first part of this perspective, I would like to highlight current developments in theoretical investigations on the structures, properties, and reactivity levels of various organic and inorganic compounds from our research group. In [1], trans- and cis-[ReCl₄(CNMe)₂] and trans- and cis-[RuCl₂(PH₃)₂(CNMe)₂] bis-isocyanide symmetrical complexes of Re(IV) and Ru(II), as well as their cycloaddition reactions with nitrone CH₂N(Me)O, were thoroughly investigated using density functional theory (DFT) techniques (CPCM-B3LYP/6-311+G(d,p)//gas-B3LYP/6-311+G(d,p) level of theory). The metal-assisted [2+3]-dipolar cycloaddition reaction of nitrones R₁CH=N(R₂)O to isocyanides CNR and the decomposition of these carbenes to imines R₁CH=NR₂ and isocyanates O=C=NR were discussed in ref. [2] (CPCM-B3LYP/6-311+G(d,p)//gas-B3LYP/6-311+G(d,p) level of theory). This reaction produces N-heterocyclic oxadiazoline carbenes which have the potential to be useful compounds for medical chemistry. In ref. [3], the driving forces for platinum(IV)-mediated nitrile–imine coupling were found to lead to the generation of asymmetrical diplatinum products that are potentially useful for medical chemistry (B3LYP/6-31G* level of theory). In ref. [4], the platinum(II)-mediated coupling mechanism between metal-activated nitriles and amidoximes was theoretically investigated at the DFT level of theory (CPCM-B3LYP/6-311+G(d,p)//gas-B3LYP/6-31G(d) level of theory). A conceptual DFT approach was used to calculate the driving forces of the hypothetical 1,3-dipolar cycloaddition (CA) of nitrone CH₂=N(Me)O to CNMe ligands and noncomplex isocyanides in various symmetric and asymmetric transition metal complexes at the CPCM-B3LYP/6-311+G(d,p)//gas-B3LYP/6-31G(d) level of theory [5]. In ref. [6], the reason for the different stability levels of nickel(II)-mediated cyano-amido-ketoxime coupling products, which are potentially useful in medicinal chemistry, was theoretically interpreted based on quantum chemical calculations at the M06-L/MDF10(Ni) and 6-31G(d) (other atoms) levels of theory. In ref. [7], based on DFT calculations at the CPCM-B3LYP/6-311+G(d,p)//gas-B3LYP/6-31G(d) level of theory, the Pd^II-mediated integration of isocyanate and azide ions was found to occur via a formal 1,3-dipolar cycloaddition between RNC ligands and uncomplexed azide. In ref. [8], interesting photophysical properties of symmetric pentaaza-nonatetraenoate (PANT) ring systems, which are potentially useful in medicinal chemistry, were discussed and analyzed using the time-dependent density functional theory (TD-DFT) at the B3LYP/LANL2DZ (Pt) and B3LYP/6-31G(d,p) (H, C, N, F, and Cl) levels of theory with dichloromethane used as the solvent environment, and a detailed analysis of the corresponding frontier molecular orbitals for the lower-lying transition was conducted. In ref. [9], various possible mechanisms associated with the nucleophilic addition of oximes to nitrile–decaborate complexes (which are potentially useful in boron neutron capture therapy in oncological diseases) were investigated via DFT calculations at the M06-2X/6-311+G(d) level of theory. In ref. [10], we observed that the reaction of cis-[PdCl₂(CNXyl)₂] (Xyl = 2,6-Me₂C₆H₃) with different 1,3,4-thiadiazole- and 1,3-thiazole-2-amines in CHCl₃ generated a mixture of two regioisomeric diaminocarbene bicyclic complexes. Results of the DFT calculations at the M06/MWB28(Pd) and 6-31G(d) (other atoms) levels of theory and the topological analysis of the electron density distribution in Bader theory formalism (QTAIM method) showed that the relative stability of the regioisomers in the CHCl₃ solution can be determined by the energy difference between the two types of intramolecular chalcogen bonds. Finally, in ref. [11], it was shown by DFT calculations at the M06/MDF10(Cu) and 6-31G(d) (other atoms) levels of theory that the cycloaddition of ketonitrones to Cu^I-linked cyanamides is a concerted process and the copper-catalyzed reaction is controlled by the dominant involvement of HOMO_dipole-LUMO_{dipolarophile} interactions. The process involving metal is more feasible than the hypothetical metal-free reaction, is both kinetically and thermodynamically profitable, and is much more asynchronous.

Otherwise, currently, big data analysis, artificial intelligence, and machine learning techniques are widely utilized for the rational and smart design of really large and complicated chemical systems (e.g., polymers and composite materials), and in the second part of this perspective, I would like to highlight current developments in these directions from other researchers. The authors of [12] discussed the recent achievements in successfully applying big data science in porous materials, particularly metal–organic frameworks (namely how to select appropriate training sets; survey approaches that are used to represent these materials in feature space; and review different learning architectures, as well as evaluation and interpretation strategies). In order to produce clean fuels, reduce the effects of global warming, and address environmental degradation, photocatalysts and electrocatalysts are essential for a sustainable future. Enhancing the efficiency of catalysts requires better photo/electrocatalytic processes and improved catalyst design methods. Research on photocatalysis and electrocatalysis may be greatly accelerated by recent developments in data science and artificial intelligence, notably through machine learning’s quick study of broad areas of materials chemistry. The utilization of machine learning for electrocatalyst and photocatalyst design and discovery, as reviewed in refs. [13,14], represents a paradigm shift in the way that advanced next-generation catalysts are designed and synthesized. Modern state-of-the-art applications of artificial intelligence and machine learning algorithms to crystallization, as analyzed in ref. [15] to present a holistic overview of machine learning and cheminformatics applications, represent a novel tool for investigating various crystal structures features and predicting their essential properties and features for smart and controllable roadmaps for crystallization dynamics processes and the future automatization of industrial valuable processes involving crystalline materials. With the development of advanced electronic devices and electric power systems, polymer-based dielectric film capacitors with high energy storage capabilities have become particularly important. Comparing polymer nanocomposites with widespread attention showed that all-organic polymers are fundamental and have been proven to be more effective choices in scalable, continuous, and large-scale industrial production, leading to many dielectric and energy storage applications, and recent progress and future prospects on all-organic polymer dielectrics for energy storage capacitors were highlighted in ref. [16]. Computational approaches for organic semiconductors, such as organic photovoltaics (OPVs), organic light-emitting diodes (OLEDs), organic thermoelectrics, organic batteries, and organic (bio)sensors (highlighting the limitations of these methods and how sophisticated physical and mathematical frameworks have been created to overcome those limitations), were discussed in ref. [17]. The discovery and optimization of materials (such as flexible solar panels, biomaterials and printable tissues, new catalysts, polymers, and porous materials) using evolutionary approaches were overviewed in ref. [18]. Recent efforts centered around applying electron microscopy to soft (including biological) nanomaterials, as discussed in ref. [19], reveal how developments of both the hardware and software of electron microscopy have enabled new insights into the formation, assembly, and functioning (e.g., energy conversion and storage, phonon/photon modulation) of these materials by providing shape, size, phase, structural, and chemical information at nanometer or higher spatial resolutions). The machine-learning-assisted identification of copolymer microstructures based on microscopic images was also discussed in ref. [20]. Improving the flame retardancy of polymeric materials used in engineering applications is an increasingly important strategy for limiting fire hazards. However, the wide variety of flame-retardant polymeric nanocomposite compositions prevents the quick identification of optimal designs for a specific application. The accelerated design of flame-retardant polymeric nanocomposites via machine learning prediction was highlighted in ref. [21]. A critical review of machine learning techniques on thermoelectric materials was presented in ref. [22]. A comparative study of machine learning approaches and response surface methodologies for optimizing the HPT processing parameters of AA6061/SiCp composites was presented in ref. [23]. The discrimination of quartz genesis based on explainable machine learning was presented in ref. [24]. Machine learning methods centered around the non-destructive testing of dynamic properties of vacuum-insulated glazing-type composite panels was discussed in ref. [25]. Mathematical modeling and computational simulation techniques were applied to a study on glycerol and/or molasses anaerobic co-digestion processes in ref. [26]. The partial decision tree forest model was highlighted as a perspective machine learning model for geosciences ref. [27]. Finally, optimizing powder metallurgy parameters to enhance the mechanical properties of Al-4Cu/xAl₂O₃ composites using machine learning and response surface approaches was reported in ref. [28].

Thus, computer modeling and machine learning in chemistry and materials science open many perspectives toward understanding various properties and reaction mechanisms (including kinetics and thermodynamics) of small organic and inorganic molecules and allow for the smart design of polymers and composites. Transition metals are widely known for their unique physical and chemical properties which often challenge theoretical modeling techniques. However, DFT techniques have been a widely used tool to study these materials, offering a reasonable description of many of their properties. Nonetheless, it is important to acknowledge that DFT techniques have limitations, particularly when they come to heavy transition metals and their complexes, due to the strong electrostatic interaction and complexity of electronic correlations in these systems. From my point of view, the application of information technologies can help researchers collect big data and understand patterns that are not obvious to humans, and this is a general perspective integral to the future of theoretical chemistry.