Latent Variable Machine Learning Framework for Catalysis: General Models, Transfer Learning, and Interpretability

Machine learning has been successfully applied in recent years to screen materials for a variety of applications. However, despite recent advances, most screening-based machine learning approaches are limited in generality and transferability, requiring new models to be created from scratch for each new application. This is particularly apparent in catalysis, where there are many possible intermediates and transition states of interest in addition to a large number of potential catalytic materials. In this work, we developed a new machine learning framework that is built on chemical principles and allows the creation of general, interpretable, reusable models. Our new architecture uses latent variables to create a set of submodels that each take on a relatively simple learning task, leading to higher data efficiency and promoting transfer learning. This architecture infuses fundamental chemical principles, such as the existence of elements as discrete entities. We show that this architecture allows for the creation of models that can be reused for many different applications, providing significant improvements in efficiency and convenience. For example, our architecture allows simultaneous prediction of adsorption energies for many adsorbates on a broad array of alloy surfaces with mean absolute errors (MAEs) around 0.20-0.25 eV. The integration of latent variables provides physical interpretability, as predictions can be explained in terms of the learned chemical environment as represented by the latent space. Further, these latent variables also serve as new feature representations, allowing efficient transfer learning. For example, new models with useful levels of accuracy can be created with less than 10 data points, including transfer learning to an experimental data set with an MAE less than 0.15 eV. Lastly, we show that our new machine learning architecture is general and robust enough to handle heterogeneous and multifidelity data sets, allowing researchers to leverage existing data sets to speed up screening using their own computational setup.

Composition Based Oxidation State Prediction of Materials Using Deep Learning Language Models

Oxidation states (OS) are the charges on atoms due to electrons gained or lost upon applying an ionic approximation to their bonds. As a fundamental property, OS has been widely used in charge-neutrality verification, crystal structure determination, and reaction estimation. Currently, only heuristic rules exist for guessing the oxidation states of a given compound with many exceptions. Recent work has developed machine learning models based on heuristic structural features for predicting the oxidation states of metal ions. However, composition-based oxidation state prediction still remains elusive so far, which has significant implications for the discovery of new materials for which the structures have not been determined. This work proposes a novel deep learning-based BERT transformer language model BERTOS for predicting the oxidation states for all elements of inorganic compounds given only their chemical composition. This model achieves 96.82% accuracy for all-element oxidation states prediction benchmarked on the cleaned ICSD dataset and achieves 97.61% accuracy for oxide materials. It is also demonstrated how it can be used to conduct large-scale screening of hypothetical material compositions for materials discovery.

Application of Machine Learning in Material Synthesis and Property Prediction

Material innovation plays a very important role in technological progress and industrial development. Traditional experimental exploration and numerical simulation often require considerable time and resources. A new approach is urgently needed to accelerate the discovery and exploration of new materials. Machine learning can greatly reduce computational costs, shorten the development cycle, and improve computational accuracy. It has become one of the most promising research approaches in the process of novel material screening and material property prediction. In recent years, machine learning has been widely used in many fields of research, such as superconductivity, thermoelectrics, photovoltaics, catalysis, and high-entropy alloys. In this review, the basic principles of machine learning are briefly outlined. Several commonly used algorithms in machine learning models and their primary applications are then introduced. The research progress of machine learning in predicting material properties and guiding material synthesis is discussed. Finally, a future outlook on machine learning in the materials science field is presented.

Insight into the Electronic Properties of Semiconductor Heterostructure Based on Machine Learning and First-Principles

A first-principles approach is a powerful means of gaining insight into the intrinsic structure and properties of materials. However, with the implementation of material genetic engineering, it is still a challenging road to discover materials with high satisfaction. One alternative is to employ machine-learning techniques to mine data and predict performance. In this present contribution, the method is taken to predict the band gap opening value of graphene in a heterostructure. First, the data of 2076 binary compounds in the Materials Project library are used to achieve visual dimensionality reduction of the data set through a t-distributed stochastic neighbor embedding (t-SNE) algorithm in unsupervised learning. Then, a series of semiconductor components are screened out and form heterostructures with graphene. Second, by means of the ensemble learning EXtreme Gradient Boost (XGBoost) algorithm and support vector machine (SVM) technology, two prediction frameworks are built to predict the band gap opening value of the graphene in the system. Finally, density functional theory (DFT) is used to calculate the energy band and density of states for comparison. Analysis shows that the prediction model has an accuracy rate of 88.3%, and there is little difference between prediction results and calculation results. We anticipate that this framework model would have fascinating applications in predicting the electronic properties of various multiphase materials.

Coacervation-Mediated Combinatorial Synthesis of Biomatrices for Stem Cell Culture and Directed Differentiation

Combinatorial screening represents a promising strategy to discover biomaterials for tailored cell culture applications. Although libraries incorporating different biochemical cues have been investigated, few simultaneously recapitulate relevant biochemical, physical, and dynamic features of the extracellular matrix (ECM). Here, a noncovalent system based on liquid-liquid phase separation (coacervation) and gelation mediated by glycosaminoglycan (GAG)-peptide interactions is reported. Multiple biomaterial libraries are generated using combinations of sulfated glycosaminoglycans and poly(ethylene glycol)-conjugated peptides. Screening these biomaterials reveals preferred biomatrices for the attachment of six cell types, including primary mesenchymal stromal cells (MSCs) and primary neural precursor cells (NPCs). Incorporation of GAGs sustains the expansion of all tested cell types comparable to standard cell culture surfaces, while osteogenic differentiation of MSC and neuronal differentiation of NPC are promoted on chondroitin and heparan biomatrices, respectively. The presented noncovalent system provides a powerful tool for developing tissue-specific ECM mimics.

Quantifying the Performance of P-Type Transparent Conducting Oxides by Experimental Methods

Screening for potential new materials with experimental and theoretical methods has led to the discovery of many promising candidate materials for p-type transparent conducting oxides. It is difficult to reliably assess a good p-type transparent conducting oxide (TCO) from limited information available at an early experimental stage. In this paper we discuss the influence of sample thickness on simple transmission measurements and how the sample thickness can skew the commonly used figure of merit of TCOs and their estimated band gap. We discuss this using copper-deficient CuCrO2 as an example, as it was already shown to be a good p-type TCO grown at low temperatures. We outline a modified figure of merit reducing thickness-dependent errors, as well as how modern ab initio screening methods can be used to augment experimental methods to assess new materials for potential applications as p-type TCOs, p-channel transparent thin film transistors, and selective contacts in solar cells.

Sol-Gel-Derived Biohybrid Materials Incorporating Long-Chain DNA Aptamers

Sol-gel-derived bio/inorganic hybrid materials have been examined for diverse applications, including biosensing, affinity chromatography and drug discovery. However, such materials have mostly been restricted to the interaction between entrapped biorecognition elements and small molecules, owing to the requirement for nanometer-scale mesopores in the matrix to retain entrapped biorecognition elements. Herein, we report on a new class of macroporous bio/inorganic hybrids, engineered through a high-throughput materials screening approach, that entrap micron-sized concatemeric DNA aptamers. We demonstrate that the entrapment of these long-chain DNA aptamers allows their retention within the macropores of the silica material, so that aptamers can interact with high molecular weight targets such as proteins. Our approach overcomes the major limitation of previous sol-gel-derived biohybrid materials by enabling molecular recognition for targets beyond small molecules.