Koji Yasuda

Chemical reaction outcome prediction presents a fundamental challenge in synthetic chemistry. Most existing machine learning (ML) approaches focus on chemical reactions of typical elements. We developed a simple ML model focused on organo‐transition metal‐catalyzed reactions (OMCRs). Instead of overall reactions observed in experiments, we let the ML model learn the sequence of simplified elementary reactions. This drastically reduced the complexity of the model and helped it find common patterns from distinct reactions. We let a graph neural network learn the reactivity index of a pair of atoms. The model was able to learn a wide variety of OMCRs, and the accuracy of reaction prediction reached 97%, even though the model has extremely fewer learnable parameters than other standards. The learned reactivity indices of bonds nicely summarize the knowledge of reactions in the dataset.

Organic friction modifiers (OFMs) added to lubricating oils to reduce friction and wear are crucial for reducing energy loss and CO₂ emissions. In our previous studies, we have developed N-(2,2,6,6-tetramethyl-1-oxyl-4-piperidinyl)dodecaneamide, referred to as C₁₂TEMPO, as a new type of OFM and experimentally demonstrated its superior performance to conventional OFMs of stearic acid and glycerol monooleate. However, the behavior of C₁₂TEMPO adsorbing onto solid surfaces from base oil during sliding, which largely dictates the lubrication performance, is yet to be elucidated. Here, we performed molecular dynamics simulations for confined shear of a C₁₂TEMPO solution in poly-α-olefin between hematite surfaces. Unlike conventional OFMs, which typically have one functional group or multiple functional groups of the same type, C₁₂TEMPO features two functional groups of different types: one amide and one terminal free oxygen radical. The results showed that adsorbed boundary films with a double-layer structure form stably during sliding, owing to double- or single-site surface adsorption and interlayer hydrogen bonding via the two functional groups. Additionally, some molecules in each of the first and second layers also form intralayer hydrogen bonding. Such multitype adsorption is unique and favorable for enhancing the strength of boundary films to withstand heavily loaded and prolonged sliding. The velocity distribution indicates that the first and second layers are solid- and liquid-like, respectively. The second layer could act as a buffer for the first layer, which is the last barrier to prevent solid–solid contact, against shear. We also found that the second layer can act as a reservoir to rapidly repair the once depleted region in the first layer because of the interlayer hydrogen bonding. The combination of the high strength and self-repair ability of the C₁₂TEMPO boundary films can rationally explain the experimentally observed properties of high load-carrying capacity, excellent antiwear effect, and high stability of friction over time.

深層学習による反応予測は，人が識別パターンを設計する必要がないため最近注目されている．しかし有機遷移金属反応は一見複雑で，深層学習の適応例は殆どなかった．我々は，論文や特許などから集めた数万件の実験反応を，素反応に分解したデータベースを構築し，グラフニューラルネットワークを用いて学習させた．最高で97.1%の精度で反応が予測できた．

It is known that each powerset quantale is embeddable into some relational unital quantale whose underlying set is the powerset of some preorder. An aim of this paper is to understand the relational embedding as a relationship between quantales and preorders. For that, this paper introduces the notion of weak preorders, a functor from the category of weak preorders to the category of partial semigroups, and a functor from the category of partial semigroups to the category of quantales and lax homomorphisms. By using these two functors, this paper shows a correspondence among four classes of weak preorders (including the class of ordinary preorders), four classes of partial semigroups, and four classes of quantales. As a corollary of the correspondence, we can understand the relational embedding map as a natural transformation between functors onto certain category of quantales.

The measurement apparatus proposed in the titled paper, "Energy-Time Uncertainty Relations in Quantum Measurements" [Found. Phys. 46, 1522-1550 (2016)] was examined. A simple proof was presented for the non-existence of the apparatus.

Herein, the ab initio method is applied to examine metastable molecular excited states on a solid surface using resonance state theory and Green’s function. A formula for the complex energy correction that determines the decay rate is presented; the configuration interaction effect together with major molecule—surface interactions are considered in more detail as compared to previous studies. Furthermore, the lifetimes of the excited states of Ru–terpyridine dyes adsorbed on an anatase surface are calculated, and the effects of the molecular structure and adsorption mode on the electron injection rate are studied. Also, the adsorption structures and relative stabilities of a series of Ru–terpyridine dyes—including the black dye—are reported. An implicit solvation model is necessary to reliably calculate the alignment between the photoabsorption spectrum and the conduction band density of states, governing the injection rate. Finally, some of the factors that limit the injection ability of dyes are discussed.

Machine learning for predicting properties of molecules and designing molecules with desired properties requires suitable numerical representation of molecular structure. In this study we propose to represent a molecular graph as a high dimensional feature vector by counting the number of substructures of a molecule, and report conditions that these vectors should satisfy. The vectors are expanded by the lattice basis extracted from the feature vectors of a large molecular dataset. In addition, an efficient algorithm to reconstruct molecular structures from the feature vector is presented. The latent space provides a continuous and concise representation of molecular graphs, and the representation allows smooth interpolation between two molecules. The proposed method has wide application to studies of quantitative structure activity relationships and machine learning of molecules.

Preoperative chemoradiotherapy has been performed as a standard therapy for advanced low rectal cancer. Cancer stem cells (CSCs) have been reported to contribute to resistance to treatment and patient prognosis. Leucine‑rich repeat‑containing G protein‑coupled receptor 5 (LGR5) and cluster of differentiation (CD133) are putative markers for CSCs. However, their prognostic ability remains unknown, and evaluation of a single marker can be insufficient due to the heterogeneity of cancer. LGR5 and CD133 expression was immunohistochemically evaluated in surgical specimens of 56 patients who received curative resection following chemoradiotherapy for advanced low rectal cancer. In addition, the correlations between their expression levels, and clinicopathological features and patient prognosis were asessed. LGR5 expression was significantly correlated with lymphatic invasion, lymph node metastasis, and tumor node metastasic (TNM) stage. CD133 expression was significantly correlated with vascular invasion and the tumor regression grade. Combined expression was significantly correlated with lymphatic invasion, tumor regression grade and TNM stage, but not with overall, and disease‑free survival. LGR5 and CD133 expressions may represent useful markers associated with tumor progression and resistance to chemoradiotherapy in patients with low rectal cancer. Furthermore, combined expression of these markers may be a more useful marker compared with the expression of each single marker.

To clarify the surgical outcomes and risk factors for anastomotic leakage (AL) following laparoscopic anterior resection (Lap-AR) for the treatment of rectal cancer. We retrospectively reviewed the records of 175 consecutive primary rectal cancer patients who had undergone Lap-AR at our institution between April 2012 and November 2015. Patient, tumor, and surgical variables were analyzed using univariate analyses. Of 175 patients, 116 were men (66.3%). All four patients who had AL (2.3%) were men and current smokers with heavy smoking histories. In three of the AL cases, preoperative total colonoscopy was impossible owing to tumor obstruction, and the other case had concomitant obstructive colitis after oral bowel preparation. Univariate analysis identified tumor size, tumor obstruction, and smoking history as factors significantly associated with AL development. Tumor size, tumor obstruction, and smoking history were risk factors for AL following Lap-AR for the treatment of primary rectal cancer.