Recent practical application of automated experiments using robotics, high-throughput experiments, and artificial intelligence technology has been progressing rapidly. In automated experiments, molecular identification is an important process for obtaining structural information on synthesized compounds and understanding their reactivity and chemical properties. In this study, we developed a system for automated molecular identification. The system uses spectral information and quantum chemical calculations, which provide no fluctuating data and have a potential to explore a wide range of chemical space. Numerical validation results suggested that the system is capable of efficient and accurate automated molecular identification in organic compounds with low molecular weight.

We have been promoting a project to evaluate the set of effective interaction parameters used in dissipative particle dynamics (DPD) simulations for all amino acid residues covering various proteins, based on fragment molecular orbital (FMO) calculations. This simulation protocol has been termed FMO-DPD. Here we report a test application to the folding problem of Chignolin and Superchignolin with hairpin structures, where 7 amino acid residues were considered.

The aggregation of polyglutamine (polyQ) proteins, which have the abnormal expansion of glutamine repeats, is a critical pathological hallmark of polyQ diseases. Experimental studies have shown an amino acid arginine uniquely inhibits the polyQ-protein aggregation. We performed replica-permutation molecular dynamics simulations to clarify the inhibitory effects of arginine on the polyQ-protein aggregation. We found arginine makes more contact with the polyQ protein than lysine, and this tendency of arginine likely inhibits the polyQ-protein aggregation.

Material Informatics (MI) needs large amount of data about materials. However, the cost of data extraction is very high. Therefore, chemical researchers are interested in technology to automatically extract the data from published documents such as patents. Previous technologies can extract the data from text in patents, but not tables. Therefore, we develop the data for MI extraction method from texts and tables in patents. In our evaluation, our method can reduce the time of data extraction by one-half. In the results, it can be expected that the new method can sufficiently reduce the cost of data extraction.

The molecular descriptor set suggested in electronic-structure informatics (ESI) was applied to predict activity of potential candidates of α-glucosidase inhibitors. In this study, we constructed a regression model for predicting the pIC₅₀ values of the known inhibitors registered in the ChEMBL database. The obtained regression model reasonably reproduced the experimental values. After refinement of the model, we conducted in silico screening to search for active inhibitors drugs from a natural product database, called KampoDB, for traditional Japanese medicine. We have discovered some promising compounds as potential α-glucosidase inhibitors.

We propose a quantum dynamics simulator for light-irradiated matter. As the coherent control of the quantum states of matter is an intriguing problem in realizing quantum technology, the quantum properties of light-matter complexes have attracted much attention. Theoretical calculations corresponding to individual experimental configurations are required to understand these properties, and simulation methods that flexibly adapt to various experimental configurations are required. In this paper, we propose a quantum dynamics simulation method that encompasses program modules corresponding to light sources, optical components, material systems, and the evaluation of physical properties. We demonstrate the entanglement generation dynamics between two spin-chains via irradiated photons and show that a flexible and expandable quantum dynamics simulator can be constructed.

The values of the entropy of 148 small organic molecules have been estimated by machine learning with only molecular orbital energies as the explanatory variables. Out of 148 molecules，we used 104 molecules for the training set and 44 molecules for the test set. We used 139 regression methods of R/caret package for machine learning. We evaluated values by RMSE (Root Mean Squared Error) and R² (coefficient of determination). From those evaluation，xgbLinear (eXtreme Gradient Boosting) and RRFglobal (Regularized Random Forest) are considered better than other regression methods. It has been proved that the entropy can be predicted by the molecular orbital energies only.

Octanol/water partition coefficient, log P, is an important parameter in classical QSAR. The new method using machine learning which we propose uses only the molecular orbital energy as an explanatory variable and does not include log P. Therefore, since the log P value can be predicted using the molecular orbital energy, we speculated that log P may not be necessary as a result if sufficient number of molecular orbital energies would be given as parameters.

The recent advances in artificial intelligence (AI) have accelerated the development of data-driven modeling. Complex machine learning models often lack interpretability. Symbolic regression, particularly in the fields of mathematics and physics, has provided alternative models that are interpretable and have excellent extrapolation capabilities. In this study, we investigated the potential of symbolic regression in chemistry, specifically in the exploration of new materials through extrapolation. We conducted fundamental verification of extrapolation and applied research on the exploration of perovskite catalysts using the recursive-LASSO-based symbolic regression. Our results suggested that symbolic regression exhibits superior extrapolation performance and interpretability compared to conventional machine learning methods.

We have derived the wavefunction and excitation energy of frozen orbital analysis (FZOA) for spin-flip time-dependent density functional theory in order to evaluate the controlling factor of the S₀/S₁ minimum energy conical intersection (MECI). The spin-complete method was applied in the derivation to avoid the spin contamination inherent in spin-flip approaches. From the results of numerical calculations, the controlling factors of S₀/S₁ MECI were found as "HOMO−LUMO exchange integral is close to zero" and "Coulomb integrals related to HOMO and LUMO determine the upper limit of the HOMO−LUMO gap". In this paper, we account for the FZOA and its formulation within the spin-fip method. Based on the application of the derived formula of spin-flip FZOA to ethylene and uracil, we explain the discovery of the controlling factors based on the excitation energy components, the electronic structure of the S₀/S₁ MECIs, and their dependence on the coupling coefficients of the restricted open-shell method.