ALK gene rearrangement was observed in 3%–5% of non-small cell lung cancer patients, and multiple ALK-tyrosine kinase inhibitors (TKIs) have been sequentially used. Multiple ALK-TKI resistance mutations have been identified from the patients, and several compound mutations, such as I1171N + F1174I or I1171N + L1198H are resistant to all the approved ALK-TKIs. In this study, we found that gilteritinib has an inhibitory effect on ALK-TKI–resistant single mutants and I1171N compound mutants in vitro and in vivo. Surprisingly, EML4-ALK I1171N + F1174I compound mutant-expressing tumors were not completely shrunk but regrew within a short period of time after alectinib or lorlatinib treatment. However, the relapsed tumor was markedly shrunk after switching to the gilteritinib in vivo model. In addition, gilteritinib was effective against NTRK-rearranged cancers including entrectinib-resistant NTRK1 G667C-mutant and ROS1 fusion-positive cancer.

Recently, molecular generation models based on deep learning have attracted significant attention in drug discovery. However, most existing molecular generation models have a serious limitation in the context of drug design wherein they do not sufficiently consider the effect of the three-dimensional (3D) structure of the target protein in the generation process. In this study, we developed a new deep learning-based molecular generator, SBMolGen, that integrates a recurrent neural network, a Monte Carlo tree search, and docking simulations. The results of an evaluation using four target proteins (two kinases and two G protein-coupled receptors) showed that the generated molecules had a better binding affinity score (docking score) than the known active compounds, and they possessed a broader chemical space distribution. SBMolGen not only generates novel binding active molecules but also presents 3D docking poses with target proteins, which will be useful in subsequent drug design. File list (2)download file view on ChemRxiv SBMolGen_manuscript_ChemRxiv.pdf (13.43 MiB) download file view on ChemRxiv SBMolGen_manuscript_Supporting_Information_ChemR...

The replacement of standard molecular mechanics force fields by inexpensive molecular orbital (QM') methods in multi-scale models has many advantages, e.g., a more straightforward description of mutual polarization and charge transfer between layers. The ONIOM(QM:QM') scheme with mechanical embedding can combine any two methods without prior parameterization or significant coding effort. In this scheme the environmental effect is evaluated fully at the QM' level and the accuracy therefore depends on how well the low-level QM' method describes the changes in electron density of the reacting region. To examine the applicability of the QM:QM' approach we perform case studies with density- Polarization effects are fairly well described using DFTB, but in some cases QM and QM' methods converge to different electronic states. We discuss when the QM:QM' approach is appropriate and the possibilities of estimating the quality of the ONIOM extension without having to make explicit benchmarks of the entire system. 3

The deposition of Amyloid-beta peptides (Aβ) is detected at an earlier stage in Alzheimer’s disease (AD) pathology. Thus, the approach toward Aβ metabolism is considered to play a critical role in the onset and progression of AD. Mounting evidence suggests that lifestyle-related diseases are closely associated with AD, and exercise is especially linked to the prevention and the delayed progression of AD. We previously showed that exercise is more effective than diet control against Aβ pathology and cognitive deficit in AD mice fed a high-fat diet; however, the underlying molecular mechanisms remain poorly understood. On the other hand, a report suggested that exercise induced expression of fibronectin type III domain-containing protein 5 (FNDC5) in the hippocampus of mice through PGC1α pathway. Thus, in the current study, we investigated a possibility that FNDC5 interacts with amyloid precursor protein (APP) and affects Aβ metabolism. As a result, for the first time ever, we found the interaction between FNDC5 and APP, and forced expression of FNDC5 significantly decreased levels of both Aβ40 and Aβ42 secreted in the media. Taken together, our results indicate that FNDC5 significantly affects β-cleavage of APP via the interaction with APP, finally regulating Aβ levels. A deeper understanding of the mechanisms by which the interaction between APP and FNDC5 may affect Aβ production in an exercise-dependent manner would provide new preventive strategies against the development of AD.Electronic supplementary materialThe online version of this article (10.1186/s13041-018-0401-8) contains supplementary material, which is available to authorized users.

Recently, molecular generation models based on deep learning have attracted significant attention in drug discovery. However, most existing molecular generation models have a serious limitation in the context of drug design wherein they do not sufficiently consider the effect of the three-dimensional (3D) structure of the target protein in the generation process. In this study, we developed a new deep learning-based molecular generator, SBMolGen, that integrates a recurrent neural network, a Monte Carlo tree search, and docking simulations. The results of an evaluation using four target proteins (two kinases and two G protein-coupled receptors) showed that the generated molecules had a better binding affinity score (docking score) than the known active compounds, and they possessed a broader chemical space distribution. SBMolGen not only generates novel binding active molecules but also presents 3D docking poses with target proteins, which will be useful in subsequent drug design.

Recently, molecular generation models based on deep learning have attracted significant attention in drug discovery. However, most existing molecular generation models have a serious limitation in the context of drug design wherein they do not sufficiently consider the effect of the three-dimensional (3D) structure of the target protein in the generation process. In this study, we developed a new deep learning-based molecular generator, SBMolGen, that integrates a recurrent neural network, a Monte Carlo tree search, and docking simulations. The results of an evaluation using four target proteins (two kinases and two G protein-coupled receptors) showed that the generated molecules had a better binding affinity score (docking score) than the known active compounds, and they possessed a broader chemical space distribution. SBMolGen not only generates novel binding active molecules but also presents 3D docking poses with target proteins, which will be useful in subsequent drug design.