Ryosuke Kojima

Bio: Ryosuke Kojima is an academic researcher from Kyoto University. The author has contributed to research in topics: Microphone array & Acoustic source localization. The author has an hindex of 12, co-authored 32 publications receiving 424 citations. Previous affiliations of Ryosuke Kojima include Kanazawa University & Honda.

Computer-aided diagnosis of lung nodule using gradient tree boosting and Bayesian optimization.

Abstract: We aimed to evaluate a computer-aided diagnosis (CADx) system for lung nodule classification focussing on (i) usefulness of the conventional CADx system (hand-crafted imaging feature + machine learning algorithm), (ii) comparison between support vector machine (SVM) and gradient tree boosting (XGBoost) as machine learning algorithms, and (iii) effectiveness of parameter optimization using Bayesian optimization and random search. Data on 99 lung nodules (62 lung cancers and 37 benign lung nodules) were included from public databases of CT images. A variant of the local binary pattern was used for calculating a feature vector. SVM or XGBoost was trained using the feature vector and its corresponding label. Tree Parzen Estimator (TPE) was used as Bayesian optimization for parameters of SVM and XGBoost. Random search was done for comparison with TPE. Leave-one-out cross-validation was used for optimizing and evaluating the performance of our CADx system. Performance was evaluated using area under the curve (AUC) of receiver operating characteristic analysis. AUC was calculated 10 times, and its average was obtained. The best averaged AUC of SVM and XGBoost was 0.850 and 0.896, respectively; both were obtained using TPE. XGBoost was generally superior to SVM. Optimal parameters for achieving high AUC were obtained with fewer numbers of trials when using TPE, compared with random search. Bayesian optimization of SVM and XGBoost parameters was more efficient than random search. Based on observer study, AUC values of two board-certified radiologists were 0.898 and 0.822. The results show that diagnostic accuracy of our CADx system was comparable to that of radiologists with respect to classifying lung nodules.

Separation of Linear and Branched Alkanes Using Host-Guest Complexation of Cyclic and Branched Alkane Vapors by Crystal State Pillar[6]arene.

Abstract: Activated crystals of pillar[6]arene produced by removing the solvent upon heating were able to take up branched and cyclic alkane vapors as a consequence of their gate-opening behavior. The uptake of branched and cyclic alkane vapors by the activated crystals of pillar[6]arene induced a crystal transformation to form one-dimensional channel structures. However, the activated crystals of pillar[6]arene hardly took up linear alkane vapors because the cavity size of pillar[6]arene is too large to form stable complexes with linear alkanes. This shape-selective uptake behavior of pillar[6]arene was further utilized for improving the research octane number of an alkane mixture of isooctane and n-heptane: interestingly, the research octane number was dramatically improved from a low research octane number (17 %) to a high research octane number (>99 %) using the activated crystals of pillar[6]arene.

kGCN: a graph-based deep learning framework for chemical structures

Abstract: Deep learning is developing as an important technology to perform various tasks in cheminformatics. In particular, graph convolutional neural networks (GCNs) have been reported to perform well in many types of prediction tasks related to molecules. Although GCN exhibits considerable potential in various applications, appropriate utilization of this resource for obtaining reasonable and reliable prediction results requires thorough understanding of GCN and programming. To leverage the power of GCN to benefit various users from chemists to cheminformaticians, an open-source GCN tool, kGCN, is introduced. To support the users with various levels of programming skills, kGCN includes three interfaces: a graphical user interface (GUI) employing KNIME for users with limited programming skills such as chemists, as well as command-line and Python library interfaces for users with advanced programming skills such as cheminformaticians. To support the three steps required for building a prediction model, i.e., pre-processing, model tuning, and interpretation of results, kGCN includes functions of typical pre-processing, Bayesian optimization for automatic model tuning, and visualization of the atomic contribution to prediction for interpretation of results. kGCN supports three types of approaches, single-task, multi-task, and multi-modal predictions. The prediction of compound-protein interaction for four matrixmetalloproteases, MMP-3, -9, -12 and -13, in the inhibition assays is performed as a representative case study using kGCN. Additionally, kGCN provides the visualization of atomic contributions to the prediction. Such visualization is useful for the validation of the prediction models and the design of molecules based on the prediction model, realizing “explainable AI” for understanding the factors affecting AI prediction. kGCN is available at https://github.com/clinfo.

Fast Tree: Computing Large Minimum-Evolution Trees with Profiles instead of a Distance Matrix

Abstract: Gene families are growing rapidly, but standard methods for inferring phylogenies do not scale to alignments with over 10,000 sequences. We present FastTree, a method for constructing large phylogenies and for estimating their reliability. Instead of storing a distance matrix, FastTree stores sequence profiles of internal nodes in the tree. FastTree uses these profiles to implement neighbor-joining and uses heuristics to quickly identify candidate joins. FastTree then uses nearest-neighbor interchanges to reduce the length of the tree. For an alignment with N sequences, L sites, and a different characters, a distance matrix requires O(N^2) space and O(N^2 L) time, but FastTree requires just O( NLa + N sqrt(N) ) memory and O( N sqrt(N) log(N) L a ) time. To estimate the tree's reliability, FastTree uses local bootstrapping, which gives another 100-fold speedup over a distance matrix. For example, FastTree computed a tree and support values for 158,022 distinct 16S ribosomal RNAs in 17 hours and 2.4 gigabytes of memory. Just computing pairwise Jukes-Cantor distances and storing them, without inferring a tree or bootstrapping, would require 17 hours and 50 gigabytes of memory. In simulations, FastTree was slightly more accurate than neighbor joining, BIONJ, or FastME; on genuine alignments, FastTree's topologies had higher likelihoods. FastTree is available at http://microbesonline.org/fasttree.

Modular and predictable assembly of porous organic molecular crystals

Abstract: Nanoporous molecular frameworks are important in applications such as separation, storage and catalysis. Empirical rules exist for their assembly but it is still challenging to place and segregate functionality in three-dimensional porous solids in a predictable way. Indeed, recent studies of mixed crystalline frameworks suggest a preference for the statistical distribution of functionalities throughout the pores rather than, for example, the functional group localization found in the reactive sites of enzymes. This is a potential limitation for ‘one-pot’ chemical syntheses of porous frameworks from simple starting materials. An alternative strategy is to prepare porous solids from synthetically preorganized molecular pores. In principle, functional organic pore modules could be covalently prefabricated and then assembled to produce materials with specific properties. However, this vision of mix-and-match assembly is far from being realized, not least because of the challenge in reliably predicting three-dimensional structures for molecular crystals, which lack the strong directional bonding found in networks. Here we show that highly porous crystalline solids can be produced by mixing different organic cage modules that self-assemble by means of chiral recognition. The structures of the resulting materials can be predicted computationally, allowing in silico materials design strategies. The constituent pore modules are synthesized in high yields on gram scales in a one-step reaction. Assembly of the porous co-crystals is as simple as combining the modules in solution and removing the solvent. In some cases, the chiral recognition between modules can be exploited to produce porous organic nanoparticles. We show that the method is valid for four different cage modules and can in principle be generalized in a computationally predictable manner based on a lock-and-key assembly between modules.

EPA-ng: Massively Parallel Evolutionary Placement of Genetic Sequences.

Abstract: Next generation sequencing (NGS) technologies have led to a ubiquity of molecular sequence data. This data avalanche is particularly challenging in metagenetics, which focuses on taxonomic identification of sequences obtained from diverse microbial environments. Phylogenetic placement methods determine how these sequences fit into an evolutionary context. Previous implementations of phylogenetic placement algorithms, such as the evolutionary placement algorithm (EPA) included in RAxML, or PPLACER, are being increasingly used for this purpose. However, due to the steady progress in NGS technologies, the current implementations face substantial scalability limitations. Herein, we present EPA-NG, a complete reimplementation of the EPA that is substantially faster, offers a distributed memory parallelization, and integrates concepts from both, RAxML-EPA and PPLACER. EPA-NG can be executed on standard shared memory, as well as on distributed memory systems (e.g., computing clusters). To demonstrate the scalability of EPA-NG, we placed $1$ billion metagenetic reads from the Tara Oceans Project onto a reference tree with 3748 taxa in just under $7$ h, using 2048 cores. Our performance assessment shows that EPA-NG outperforms RAxML-EPA and PPLACER by up to a factor of $30$ in sequential execution mode, while attaining comparable parallel efficiency on shared memory systems. We further show that the distributed memory parallelization of EPA-NG scales well up to 2048 cores. EPA-NG is available under the AGPLv3 license: https://github.com/Pbdas/epa-ng.

Nonporous Adaptive Crystals of Pillararenes

Abstract: ConspectusPorous materials with high surface areas have drawn more and more attention in recent years because of their wide applications in physical adsorption and energy-efficient adsorptive separation processes. Most of the reported porous materials are macromolecular porous materials, such as zeolites, metal–organic frameworks (MOFs), or porous coordination polymers (PCPs), and porous organic polymers (POPs) or covalent organic frameworks (COFs), in which the building blocks are linked together by covalent or coordinative bonds. These materials are barely soluble and thus are not solution-processable. Furthermore, the relatively low chemical, moisture, and thermal stability of most MOFs and COFs cannot be neglected. On the other hand, molecular porous materials such as porous organic cages (POCs), which have been developed very recently, also show promising applications in adsorption and separation processes. They can be soluble in organic solvents, making them solution-processable materials. However, ...

A Review of Deep Learning Methods and Applications for Unmanned Aerial Vehicles

Abstract: Deep learning is recently showing outstanding results for solving a wide variety of robotic tasks in the areas of perception, planning, localization, and control. Its excellent capabilities for learning representations from the complex data acquired in real environments make it extremely suitable for many kinds of autonomous robotic applications. In parallel, Unmanned Aerial Vehicles (UAVs) are currently being extensively applied for several types of civilian tasks in applications going from security, surveillance, and disaster rescue to parcel delivery or warehouse management. In this paper, a thorough review has been performed on recent reported uses and applications of deep learning for UAVs, including the most relevant developments as well as their performances and limitations. In addition, a detailed explanation of the main deep learning techniques is provided. We conclude with a description of the main challenges for the application of deep learning for UAV-based solutions.