Artificial intelligence chemistry最新文献

Machine learning (ML) techniques are already widely and increasingly used in diverse applications in science and technology, including computational chemistry. Specifically in computational chemistry, neural networks (NN) and kernel methods such as Gaussian process regressions (GPR) have been increasingly used for the construction of potential functions and functionals for density functional theory. While ML techniques have a number of advantages vs intuition-based models, notably their generality and black-box nature, they are still challenged when faced with high dimensionality of the feature space or low and uneven data density – in part because of their general nature. We review recent works using methods such as NNs and GPR as building blocks of composite methods in the framework of an expansion over orders of coupling. We introduce models using NN or GPR-based components as part of HDMR (high-dimensional model representations)-based structures. HDMR is a formalization of orders-of-coupling representations that include the many-body and N-mode representations well known in computational chemistry and allows, in particular, building all terms from one dataset of arbitrarily distributed data. The resulting HDMR-NN and HDMR-GPR combinations and NN with HDMR-GPR derived neuron activation functions not requiring non-linear optimization enhance machine learning capabilities in high dimensional spaces and or with sparse data.

The proposed work uses centrality measures based heuristic method to improve the efficiency of the solution for the similarity search problem in molecular chemical graphs by effectively identifying central candidate or representative candidate nodes, which simplify the complex processes involved while detecting a large-sized maximal common connected edge subgraph. After analyzing the structure of the two input molecular chemical graphs, a Tensor Product graph is created. This newly built graph is further analyzed to get the similarity pattern of the input graphs. It is an open problem to decide which centrality measure selects the best central candidate node in Tensor Product graphs to get a large maximal common connected edge graph. Since each centrality measure is analyses, the given graph is uniquely based on its own specific aspects. The proposed work offers directions on using various centrality measures to determine a big-sized maximal common connected subgraph for two molecular chemical input graphs. It also analyses seven centrality measures to select the best candidate node in the Tensor Product graph of two input chemical molecular graphs. Based on the obtained results, the betweenness centrality and degree centrality measures exclusively help to get large-sized similarity patterns.

The benefits of using machine learning approaches in the design, optimisation and understanding of homogeneous catalytic processes are being increasingly realised. We focus on the understanding and implementation of key concepts, which serve as conduits to more advanced chemical machine learning literature, much of which is (presently) outside the area of homogeneous catalysis. Potential pitfalls in the ‘workflow’ procedures needed in the machine learning process are identified and all the examples provided are in a chemical sciences context, including several from ‘real world’ catalyst systems. Finally, potential areas of expansion and impact for machine learning in homogeneous catalysis in the future are considered.

Neural Network (NN) models provide potential to speed up the drug discovery process and reduce its failure rates. The success of NN models requires uncertainty quantification (UQ) as drug discovery explores chemical space beyond the training data distribution. Standard NN models do not provide uncertainty information. Some methods require changing the NN architecture or training procedure, limiting the selection of NN models. Moreover, predictive uncertainty can come from different sources. It is important to have the ability to separately model different types of predictive uncertainty, as the model can take assorted actions depending on the source of uncertainty. In this paper, we examine UQ methods that estimate different sources of predictive uncertainty for NN models aiming at protein-ligand binding prediction. We use our prior knowledge on chemical compounds to design the experiments. By utilizing a visualization method we create non-overlapping and chemically diverse partitions from a collection of chemical compounds. These partitions are used as training and test set splits to explore NN model uncertainty. We demonstrate how the uncertainties estimated by the selected methods describe different sources of uncertainty under different partitions and featurization schemes and the relationship to prediction error.

A reverse molecule contribution (reMC) - molecule contribution (MC) – Machine learning (ML) protocol for disassemble and reproduce the spectrum is presented. By splitting the mixture spectrum with monochromophoric spectra in the database in a “Peeling-Onion” manner, a new recipe can be obtained. Upon comparison of the reproduced spectrum (with the forward molecular contribution - machine learning method) with the original one, the reliability of the proposed method is justified.

We present a machine learning framework for modeling the absorption properties of azobenzene molecules – an important class of organic compounds with many potential photochemical applications. The framework utilizes predictors based on the chemical composition and structure of each molecule and consists of separate regression models trained to predict the absorption at distinct wavelengths, covering the UV and visible light ranges. Despite the relatively small size of the dataset (330 molecule-absorption spectrum pairs), the models were able to learn to accurately predict the absorption at fixed wavelengths, as well as the position and intensity of the maximum absorption. These predictions can be used to rapidly screen thousands of candidate molecules for a variety of potential applications, reducing the need for time-consuming and expensive experiments or first-principles computations.

The Pd-catalyzed Buchwald–Hartwig coupling reaction is important in the construction of the C-N bond due to various applications in organic synthesis. Quantum chemical calculations are widely used in understanding reaction mechanisms whereas the machine learning method is extremely popular in recognizing the relationships of data. Here, we combine density functional theory calculations with the support vector regression method to probe the origin of the higher efficiency of terphenyl phosphine ligand over the biaryl counterpart in the Buchwald–Hartwig C-N coupling reaction. By quantum chemical calculations, the turnover frequency-determining transition states are located and ligand features are calculated with high accuracy. By machine learning, the relationship between the reaction barrier and ligand features has been examined. It is found that the interplay of the charge on the metal center, the cone angle of the ligands, and the Sterimol L parameters of the ligand determines the catalytic performance of the palladium catalysts with different phosphine ligands. Our findings could help experimental chemists to design the ligands for Pd-catalyzed C-N coupling reactions with high efficiency.