Session Computational Physics, Computational Fluid Dynamics, and Solid Mechanics
Prof. Abir De Sarkar
Institute of Nano Science and Technology, Quantum Materials & Devices Unit, Knowledge City, Sector – 81, Mohali, Punjab, India
Insights Into Applied 2D Materials For Energy Conversion And Next Generation Electronics
Several interesting properties arise at the nanoscale which are non-existent in the bulk form of these materials. For instance, bulk or multilayer van der Waals solid MoS2 does not show any piezoelectricity, while monolayer MoS2 is piezoelectric. Piezoelectricity at the nanoscale is relatively new and in its infancy. Piezoelectric properties have been explored in a few selected semiconducting 2D materials in our work using DFT based approaches. The feasibility in the synthesis of Janus MoSSe monolayer has motivated us to investigate piezoelectricity in Janus structures in centrosymmetric Group IV transition metal dichalcogenide and trichalcogenide monolayers. Ultrahigh out-of-plane piezoelectricity is found to meet giant Rashba effect in 2D Janus monolayers and bilayers of Group IV transition metal trichalcogenides. The latter is desired in spintronics. Out-of-plane piezoelectricity is found to be induced at the interfaces of semiconducting GaN, MoS2 and boron monophosphide (BP) monolayers which show σh symmetry and in-plane piezoelectricity only. These 2D semiconducting monolayers have been synergistically combined to constitute van der Waals heterobilayers (vdWH) of GaN/BP and MoS2/BP. Again, MoSSe/BP vdWH shows 4-fold enhancement in out-of-plane piezoelectricity with respect to that of MoS2/BP, which has been later supported by experimental findings. The band gap closure in MoSSe/BP via vertical electric field showcases its immense application prospects in digital electronics. Conflux of tunable Rashba effect and piezoelectricity has been observed in flexible MgTe, CdTe and ZnTe monolayers, which signifies its superhigh potential for applications in next-generation spintronic devices, e.g., in self-powered flexible-piezo-spintronics. HfN2 monolayer is found to exhibit valleytronic properties complementary to that in single-layer MoS2, while merger of spintronic with valleytronic properties are noted in h-NbN and h-TaN monolayers.
Assoc. Prof. Worasak Sukkabot
Department of Physics, Faculty of Science, Ubon Ratchathani University, 85 Sathollmark Rd. Warinchamrab, Ubon Ratchathani, Thailand, 34190
Shape-Dependent Core/Shell Nanocrystals With Interesting Electronic And Optical Properties: Atomistic Tight-Binding Theory
The electronic and optical signatures of the shape-dependent core/shell nanocrystals are determined by the atomistic tight-binding theory and configuration interaction method. Here, CdSe/CdS core/shell dot-in-rod nanocrystals and dot-in-hexagonal platelet nanocrystals are the shape-dependent models. The resulting calculations are mainly sensitive with their shapes and sizes. The quantum confinement effect is attributed to the reduction of excitonic band gaps with the increasing sizes. The important findings by the tight-binding model agree well with the experimental data. Finally, this theoretical study delivers the important direction to predesign of core/shell nanocrystals with shape-dependence and specific characteristics for the optoelectronic technologies.
Assoc. Prof. Adisak Boonchun
Department of Physics, Faculty of Science, Kasetsart University, Bangkok 10900, Thailand
Pentagonal Two- Dimensional Materials: A Promising Anode Materials For Li-Ion Battery
Two-dimensional (2D) materials with penta-atomic-configuration such as penta-graphene, penta-B2C and penta-BN2 have received great attention as an anode in Li-ion batteries (LIBs). In particular, penta-graphene, penta-B2C and penta-BN2 possess the large storage capacity for LIBs of 1489, 1594, and 2071mAh g-1 respectively. Recently, penta-BCN has been demonstrated to exhibit the highest theoretical capacity to date of 2183 mAh g-1, corresponding to the composition Li3BCN. Herein, we study the layer-by-layer Li adsorption on penta-BCN by explicitly and comprehensively considering its anisotropic structure. We discover a new, more energetically favorable Li adsorption site distinct from the latest report by Chen et al. [Chen et al. PCCP, 2021, 23, 17693] The possible migration pathway and the accompanying activation energy are also investigated. The full lithium adsorption leads to the formula Li2BCN and reduces theoretical capacity to1455 mAh g-1. Still, the penta-BCN exhibits metallic conductivity during Li adsorption, has a low open-circuit voltage, and has a low ion diffusion barrier, which are all beneficial for anode materials. These observations imply that penta-BCN remains one of the most effective anode materials for LIBs with a quick charge/discharge rate.
Asst. Prof. Suraphong Yuma
Department of Physics, Faculty of Science, Mahidol University, Bangkok, Thailand
Improving Systematic Search Of Large-Scale Outflowing Galaxies With Subaru/Hyper Suprime-Cam Survey
Galactic-scale outflow is a crucial process in halting star forming process in a galaxy and enriching the chemical abundance of the universe. However, searching for galaxies in the large-scale outflowing phase is difficult. The Hyper Suprime-Cam (HSC) Subaru strategic program is a deep imaging survey of over 1400 square degree of the sky with Subaru telescope. It provides incredibly large public data containing more than million galaxies at various epochs of the universe. Using the public HSC data, we discover more than 800 galaxy candidates with strong emission line spatially extended over 30 kiloparsecs beyond the stellar components of a typical galaxy. The number of candidates is 10 times more than those from our last search in 2017. In this talk, we will focus on the newly improved selection method including simulating artificial objects with outflow to evaluate the reliability of the search.
Asst. Prof. Nongnuch Artrith
Materials Chemistry and Catalysis, Debye Institute for Nanomaterials Science, Utrecht University,
Development of Efficient and Accurate Machine-Learning Potentials For the Simulation of Complex Molecules And Materials
The properties of materials for energy applications, such as heterogeneous catalysts and battery materials, often depend on complicated chemical compositions and complex structural features including defects and disorder. This complexity makes the direct modelling with first principles methods challenging. Machine-learning (ML) potentials trained on first principles reference data enable linear-scaling atomistic simulations with an accuracy that is close to the reference method at a fraction of the computational cost. ML models can also be trained to predict the outcome of simulations or experiments, bypassing explicit atomistic modelling altogether. Here, I will give an overview of our contributions to the development of ML potentials based on artificial neural networks (ANNs) and applications of the method to challenging materials classes including metal and oxide nanoparticles, amorphous phases, and interfaces. Further, I will show how large computational and small experimental data sets can be integrated for the ML-guided discovery of catalyst materials. These examples show that the combination of first-principles calculations and ML models is a useful tool for the modelling of nanomaterials and for materials discovery. All data and models are made publicly available. To promote Open Science, we also formulated guidelines for the publication of ML models for chemistry that aim at transparency and reproducibility.
Asst. Prof. Tirawut Worrakitpoonpon
UtrechSchool of Physics, Institute of Science, Suranaree University of Technology, Nakhon Ratchasima, 30000, Thailand
The Revisit Of The Spherical Collapse Experiment With Finite Velocity Dispersion
Spherical collapse model is widely employed as a toy model to understand the origin of elliptical galaxies. In point-wise systems, the seed of triaxiality is originated from the Poissonian fluctuations before it is amplified by the collapse. However, the conclusion for systems with finite velocity dispersion is different because of the initial velocity dispersion. We propose the theoretical framework derived from the Lin-Mestel-Shu instability incorporated with the discretized effects of the density and the velocity dispersion. This yields the critical particle number N. Below this number, the symmetry breaking is effective, while it is ineffective above that number. We find that it predicts reasonably well the simulated collapse with low density gradient.
Dr. Tanveer Hussain
UniveSchool of Science and Technology, University of New England, Armidale, New South Wales 2351, Australiarsity of New England, Australia
Insights Into the Trapping Mechanism of Light Metals on C2n-Monolayer: Efficient Anode Materials for Metal Ion Batteries
This work presents insights from atomistic simulations based on density functional theory on the trapping mechanism of light metal atoms in nitrogenated holey graphene (C2N). The storage capacity of C2N for lithium, sodium and calcium ions has been explored for utilization as a battery anode material. Our calculations show that a significant capacity of over 500 mAh g−1 and excellent mobility of the calcium atoms can be achieved. Overall, we find that all three metals interact strongly with the pyridinic nitrogen in the pores and that the material shows a high initial storage capacity. However, due to the strong binding of the first intercalated metals in the pores, these show poor mobility. Once the pores are loaded with at least one metal atom, the mobility improves significantly. The trapped metal atoms however, affect the capacity of the material, making it much smaller. This limits the suitability of C2N as an anode material for lithium and sodium ion batteries and explains previous experimental findings on poor performance for lithium. For calcium we find that the trapping of some of the calcium atoms has less of an effect due to the dual valence, leading to the observed higher capacity.
Asst. Prof. Thana Sutthibutpong
Theoretical and Computational Physics Group, Department of Physics, KMUTT, Thailand. Faculty of Science, King Mongkut’s University of Technology Thonburi
Biophysical Interpretation of Protein Engineering And Evolution by Molecular Modelling and Network Topology Analysis
This talk presents the attempts for biophysical understanding on the effects of local changes to global protein structures within protein engineering and evolution, aided by molecular simulation techniques and information theory approaches. The talk will start with discussing about how protein structures hierarchy is defined into primary, secondary, and tertiary structures so that certain patterns of protein folding can be understood in terms of physical interactions. This understanding leads to a new research field, protein engineering, in which evolution of proteins at the molecular level is mimicked and accelerated by altering a few amino acids, which affecting both local physical properties and the global interaction network. After that, basic principles and theoretical backgrounds of molecular simulations and their applications in protein engineering in collaborations with experimental labs will be discussed. Finally, the current advances and future trends of the protein structure research involving new techniques, such as, machine learning and network theory will be presented.
Dr. Teeraphat Watcharatharapong
Department of Physics, Faculty of Science, Kasetsart University, Thailand
Unveiling the Non-Stoichiometric Effect on Sodium Intercalation Mechanism and Electrochemical Properties in 3.8-V Alluaudite Based Cathodes for Na-Ion Battery: A First-Principles Study
Sodium iron sulfate in the form of alluaudite Na2Fe2(SO4)3 has emerged as one of the most promising cathodes for Na-ion batteries due to its highest Fe2+/3+ redox potential, low cost, sustainability, and high-rate capability. Unlike other cathodes, this material commonly occurs in a non-stoichiometric form of Na2.5Fe1.75(SO4)3 (N2.5F1.75S) with partial Na+ substitutions on Fe sites, depending on the synthesis conditions. While many contemporary works have primarily been directed to study this non-stoichiometric compound, our theoretical prediction unveiled the possibility to synthesize stoichiometric alluaudite (N2F2S), which is expected to deliver higher specific capacity (»120 mA.h/g). This provokes curiosity toward the non-stoichiometric effect on the electrochemical activities and sodium intercalation mechanism in alluaudite materials. In this talk, the structural evolution, electrochemical behavior, and voltage profile of NxFyS (where y = 2, 1.75, and 1.5) will be present on the basis of first-principles investigations. These results reveal the likelihood of two-phase transitions after half desodiation process, which depends on a degree of non-stoichiometry and suggests an improvement in the structural reversibility for N2.5F1.75S and N3F1.5S. The prediction of the voltage profiles shows the benefit of non-stoichiometry in enhancing the specific capacity and identifies the structural rearrangement of Fe2O10 dimers as the hidden reason behind the irreversible sharp peak experimentally observed in differential galvanostatic profiles.
Session Computational Chemistry
Prof. Cheng-chau Chiu
National Sun Yat-sen University, Department of Chemistry,
No.70 Lien-hai Rd., Kaohsiung 804, Taiwan
Exploring the Fate of Sulfur-Vacancy Sites on A MoS2 Monolayer
MoS2 has been discussed as a catalyst for different reactions, including hydrogen evolution and CO2 hydrogenation. Furthermore, it is often mentioned that the defect sites on the catalyst play an essential role for the catalytic activity. [1,2] Earlier studies have pointed out that the size and shape of the defects may affect the electronic structure  and, with that, likely also the performance of the catalyst. In this work, we use periodic DFT calculations to explore the diffusion of S-vacancy sites on a MoS2 monolayer and investigate how the immediate surrounding influences the energetics of the diffusion. These data are then used as input for a kinetic Monte Carlo model that mimics the migration of the vacancy sites and, as a consequence of that, the reshaping of the catalyst surface over time. Our results show that the migration of the defect sites heavily depends on the occupation of one distinct S-site in the direct vicinity of the migrating vacancy site. This, in return, results in the fact that only “defect islands” consisting of more than three neighboring vacancy sites are mobile at temperatures below 100°C. Furthermore, the results of our kMC simulation reveal that wide, extended vacancy sites are not very stable and quickly rearrange to a structural motive that features isolated S-atoms surrounded by a ring of vacancy sites.
Prof. Mu-Jeng Cheng
Department of Chemistry, National Cheng Kung University, Tainan, Taiwan
Applications of Quantum Mechanics to Study and Design Electrocatalysts for Hydrocarbon Partial Oxidation
The electrochemical oxygen evolution reaction (OER) is a critical bottleneck in artificial photosynthesis. It requires a very positive onset potential (UOER > 1.5 VRHE), much larger than the equilibrium potential (UOER,Eq = 1.23 VRHE), to drive the reaction, and thus is very energy demanding. For most known electrocatalysts, a high UOER is used to overcome the third elementary step of the reaction (*O + H2O(l) ® *OOH + H+ + e–). If the goal is to generate *O rather than complete the OER, the required UOER would be less positive, and the whole process would be less energy demanding. In homogeneous and heterogeneous catalysis, *O often acts as a reaction center for various oxidation reactions. Thus, *O, generated during the course of the OER, should be able to use as an active site for oxidation. In the past two years, we used quantum mechanics combined with a constant electrode model to verify and apply this concept to design and screen electrocatalysts to oxidize inexpensive hydrocarbons into more valuable compounds. In this presentation, we will detail the discovery.
Prof. Ming Kang Tsai
Department of Chemistry, National Taiwan Normal University, Taipei, Taiwan
Exploring the Molecular Electronic Properties Using Knowledge-Based and Structural Information
Organic fluorescent molecules play critical roles in the fluorescence inspection, biological probes, and labeling indicators. Learning the design principle of these molecular architectures always attract the scientific interests of the synthetic and theoretical communities. In this talk, two practices of predicting molecular electronic properties will be demonstrated using 10k-plus real world experimental and 100k-plus theoretical datasets, being represented by the chemical knowledge or physical structural information. A systematic informatics procedure will be introduced, starting from descriptor cleaning, descriptor space reduction, and statistical-meaningful regression to build an applicable model for estimating the fluorescence emission wavelength. Additionally, the comparison in terms of prediction accuracy and data preparation between machine learning and neuron network approaches will be discussed.
Prof. Ranjit Thapa
Department of Physics, SRM University – AP, Amaravati 522240, Andhra Pradesh, India
Electronic Descriptor then Predictive model using QM/ML Approach then Experimental Validation: OER on Carbon Catalyst
Energy storage and conversion devices including metal–air batteries and fuel cells are highly dependent on cathodic Oxygen Reduction Reaction (ORR). The carbon–based materials are identified as potential catalysts for ORR with ideal electronic properties but the random approach using an experimental method delays the finding of the best carbon catalysts in large material space. To overcome this challenge, we proposed a predictive model equation based on QM/ML approach that can find the site–specific ORR activity of the graphene–based system. Here, we demonstrate the key role of π orbital descriptors (Dπ(EF), R–Oπ) to influence the adsorption property of carbon sites. By using these descriptors as features, we employed various ML models and observed the optimal predictive performance of the SVR method. We applied SVR based predictive model to estimate the ORR activity of unknown graphene systems and the accuracy of the results is validated by DFT calculations. Ultimately, I will discuss the validation of the prediction with experimental results (collaborative work).
Prof. Zhang Ruiqin
Department of Physics, City University of Hong Kong, Hong Kong SAR, People’s Republic of China
Δ-Machine Learning-Driven Discovery of Double Hybrid Organic-Inorganic Perovskites
Double hybrid organic-inorganic perovskites (DHOIPs) with excellent optoelectronic properties and low production costs are promising in photovoltaic applications. However, DHOIPs still have not been investigated thoroughly, due to their structural complexities. In this work, an accelerated discovery of DHOIPs has been realized by combining machine learning (ML) techniques, high-throughput screening, and density functional theory calculations. Different from the previous works, the anisotropy of organic cations of DHOIPs was first considered, and Δ-machine learning (Δ-ML), which uses low-level calculations as a baseline to predict properties of high-level methods, was used in high-throughput of DHOIPs to further improve the accuracy of ML models. 19 promising DHOIPs with appropriate bandgaps for solar cells were screened out from 78400 DHOIPs and verified by performing HSE06 calculations. This work demonstrates an effective method for predicting and discovering hidden novel photovoltaic materials.
Prof. Deva Priyakumar
Center for Computational Natural Sciences and Bioinformatics, International Institute of Information Technology, Hyderabad 500 032, India
Modern Machine Learning for Molecular Design
Modern machine learning methods have had phenomenal success in the technology areas such as computer vision, speech recognition, natural language processing (NLP), etc. Inspired by this, we see a surge in the use of artificial intelligence-machine learning methods to address problems in fundamental sciences during the last few years. Availability of high performance Graphics Processing Unit (GPU) accelerators, large datasets, novel algorithms, and libraries have positively contributed to enhanced activity in this area. In chemistry, ML methods have been successfully applied to various problems such as predicting accurate energies of molecules, various drug discovery tasks, retrosynthetic pathway prediction, inverse design of molecules, etc. This talk will discuss the impact AI methods in general have made in chemistry research and gives a conceptual overview of different ML methods. Recent efforts of using artificial neural networks, convolutional neural networks, reinforcement learning and Monte Carlo tree search for carrying out various prediction/classification tasks in molecular design, molecular generation, and prediction of molecules starting from their spectra will be discussed.
Asst. Prof. Suwit Suthirakun
School of Chemistry, Institute of Science, Suranaree University of Technology, Nakhon Ratchasima, Thailand
First-Principles Modelling of Two-Dimensional C3n4/Zno Heterostructures as Potential Photocatalysts for Water Splitting
Photoelectrochemical water splitting has received extensive attention as an approach for the clean hydrogen production. However, the performance of current photocatalysts is still not satisfied due to high charge recombination rate and sluggish kinetics of water splitting. Making heterostructure from 2D-materials is among efficient approaches to enhance the photocatalytic activity since it allows for engineering of band edge position and provides various active sites for hydrogen and oxygen evolution reactions (HER and OER). In this talk, I will present how first-principles modelling can be used to study the photocatalytic performance of 2D heterostructures. In particular, I will discuss the formation of g-C3N4/ZnO heterostructure where biaxial strained are introduced to achieve stable and effective photocatalysts. Its band edge potential with respect to the normal hydrogen electrode is constructed to describe the type of heterojunction and photocatalytic mechanisms. In addition, the calculated free energy profiles of OER and HER are used to determine the active sites of the heterostructure, estimate the overpotential, and predict the catalytic activity of the photocatalysts. The obtained insight can be used as a guidance to rationally design 2D heterostructures for photocatalytic water splitting.
Dr. Manussada Ratanasak
Institute for Catalysis, Hokkaido University, Kita 21, Nishi 10, Sapporo, Hokkaido 001-0021, Japan
DFT Study on the Chemoselective Transesterification of Methyl Acrylate with Benzyl Alcohol Catalyzed by Magnesium(II) Aryloxides
Acrylates are produced on a million-ton scale per year as some of the most important industrial chemicals. Professor Ishihara have found sodium or magnesium aryloxides can catalyze the transesterification of methyl (meth)acrylate at room temperature, with high chemoselectivity, producing a high yield of (meth)acrylate ester, and without the use of toxic metals or ligands. In the present study, we performed the DFT calculations at the wb97XD/6-31G(d,p)level for elucidation the chemoselective transesterification (1,2-addition) of methyl acrylate (MA) with benzyl alcohol (BnOH) using the Mg(II) aryloxides catalyst by compare with undesired Michael (1,4) addition reaction. Before investigating the reaction pathways, we studied the relative stability of cis-trans isomerization of MA. From the calculated energy deviation result of two isomers is so small (0.7 kcal/mol). Hence, the reaction pathways for both cis and trans MA cases of monomeric models were investigated. The potential energy profiles of monomeric models revealed that transesterification of cis-MA is the most favorable pathway (Ea1tc = 14.9 kcal/mol) among three main pathways (Ea1tt = 15.3 kcal/mol and Ea1m = 16.9 kcal/mol). In addition, the dimeric Mg(II) complex is only 1.0 kcal/mol more stable than the monomeric Mg(II) complex. Therefore, dimeric models were also investigated. Again, the apparent activation energy of the rate-determining step of dimeric model for the transesterification pathway (Ea1tc = 13.0 kcal/mol) is smaller than that of the Michael addition pathway (Ea1m = 14.4 kcal/mol). These calculation results confirmed explanation of the preference for transesterification over Michael addition.
Session Computational Biology, Bioinformatics, Biochemistry, and Biophysics
Prof. Norio Yoshida
Graduate School of Informatics, Nagoya University, Furo-cho, Chikusa-ward, Nagoya 464-8601 Japan
Statistical Mechanics Theory of Biomolecular Solvation
Biomolecules maintain their structure and express their functions in solution. Therefore, information on the solvation of biomolecules is essential for understanding biological processes. The most interesting task in protein computational science is protein structure prediction. In recent years, AI-based methods, such as alpha-fold2, have realized accurate predictions of the stable structure of proteins. On the other hand, for dynamic conformational changes such as protein denaturation, molecular simulation-based methods are still essential. Recently, we developed a computational method to investigate the global conformational change of a protein is proposed by combining the linear response path following (LRPF) method and three-dimensional reference interaction site model (3D-RISM) theory, which is referred to as the LRPF/3D-RISM method. The improvement in the model of water molecule is another important factor. For example, the electronic polarization of water molecules is known to have a significant effect on the electron transfer process in aqueous solution. Therefore, we have developed a novel 3D-RISM theory that can describe the electronic polarization of solvents, which is referred to as the solvent-polarizable (sp) 3D-RISM theory. In this talk, the fundamentals and applications of theories will be presented.
Asst. Prof. Kowit Hengphasatporn
Center for Computational Sciences, University of Tsukuba, 1-1-1 Tennodai, Tsukuba, Ibaraki 305-8577, Japan
Ligand-Binding Mode Evaluation of Potential Sars-Cov-2 Main Protease Inhibitors Using LB-PaCS-MD/FMO Technique
Rubraxanthone, the natural compound extracted from the pericarp of mangosteen (Garcinia mangostana), has been confirmed an antiviral activity to SAR-CoV2 at the main protease as a competitive inhibitor, with low cytotoxicity and high percent inhibition to SAR-CoV2 infected cells. However, the inhibitory mechanism at the atomistic level is still ambiguous. To understand how rubraxanthone interacts with the main protease, Ligand Binding Path Sampling Based on Parallel Cascade Selection Molecular Dynamics (LB-PaCS-MD) method is used to disclose the ligand’s path from the water to the active site of SAR-CoV2 main protease. The ligand-binding pattern could reveal the necessary interaction profile for optimizing the novel protease inhibitor. The potent binding conformation is identified using fragment molecular orbital (FMO) with RI-MP2/PCM method. Ten different binding patterns are clustered into two groups using correlation distance and average linkage based on their pair interaction energy. The complexes that form the better cluster are chosen to perform MD simulation until the ligand flies out of the pocket or 500 ns to evaluate the binding stability. Among all events, one of the possible binding patterns of rubraxanthone in complex with SAR-CoV2 main protease at catalytic regions has been proposed in this study. Moreover, the interaction profile suggested that the primary interaction is methionine sulfur-aromatic interactions. The constant fluctuation of rubraxanthone in the binding pocket is a key inhibition by interfering with the catalytic dyad and interrupting the main protease’s function. Overall, the integrated LB-PaCS-MD/FMO method can provide a more reasonable complex structure necessary for further antiviral drug discovery and design.
Asst. Prof. Marasri Ruengjitchatchawalya
Biotechnology/Bioinformatics and Systems Biology Programs, King Mongkut’s University of Technology Thonburi, Bangkok, Thailand
Bioactive Peptides Screening and Discovery By Bioinformatics Machine Learning Tools
Bioactive peptides, including biological sources-derived peptides with different biological activities, are protein fragments that influence the functions or conditions of organisms, in particular humans and animals. Conventional methods of identifying bioactive peptides are time-consuming and costly. To quicken the processes, Bioinformatics and Computational Biology tools are recently used to facilitate screening of the potential peptides prior their activity assessment in vitro and/or in vivo. In the presentation, the bioactive peptides derived from particular valuable organisms including Spirulina (Arthrospira) sp. and Ophiocordyceps sinensis, screened/ discovered by our efficient computational tools will be described.
Asst. Prof. Duangrudee Tanramluk
Institute of Molecular Biosciences, Mahidol University, Salaya, Nakhon Pathom, Thailand 73170 Thailand
MANORAA: A Data-Driven Drug Design System to Connect Biological And Physical World
The world is in urgent need of centralized drug design system to connect crucial biochemical data sources together with providing meaningful physical interpretation. Our machine learning drug design platform has an acronym from Mapping Analogous Nucleis onto Residues and Affinity (MANORAA), which blends the interface of the biochemical and biophysical world. For molecular design, the physical world can be understood through biomolecular interaction types, distances, and shape. The biological world can be understood through the affected biochemical pathways, functions, and their target organs. We employ machine learning to enable the understanding of the binding kinetics of protein and ligand from inter-residue distances, intermolecular interactions, and the imaginary shape based on frequency found in the pocket. In addition, our big data backend provides a missing link to bridge those ligands to target proteins, variants, pathways, tissues and organs. Our MANORAA algorithm can successfully predict the binding affinity from inter-residue distance and guide whether the pocket should be expanded or contracted to gain better binding affinity. The resulting equations were proven experimentally with X-ray crystallographic structures and kinetics data of Plasmodium falciparum Dihydrofolate Reductase-Thymidylate Synthase in complex with Trimethoprim and Methotrexate. It can also shed light to biological functions and possible target organs from thousands of ligands in the Protein Data Bank (PDB). This platform can be accessed via MANORAA.org.
Asst. Prof. Puey Ounjai
Department of Biology, Faculty of Science, Mahidol University, Thailand
Computational Investigations of the Endoplasmic Reticulum Transport Dynamics
Recent advances in computational power and cryogenic electron microscopy have revolutionized modern structural biology. A plethora of protein structural data have recently become available, which open new ventures for further understanding of variety of cellular processes at the molecular level. However, understanding the large-scale dynamics of molecular machines inside the cells is still very challenging. Here, we exploited various computational approaches including coarse-grained (CG) and atomistic molecular dynamics (MD) simulation to investigate the structural dynamics of membrane proteins of endoplasmic reticulum (ER). Here, we used CG-MD simulations to investigate two distinct cellular recognition pathways; the molecular mechanism underpinning cellular cholesterol homeostasis and recognition of substrate by Transporter associated with antigen processing (TAP). The results from our CG and atomistic MD simulation provides novel insights, yet strongly agrees with previous biophysical and biochemical studies. Together, we demonstrated the power of CG-MD simulation in the understanding of a larger scale molecular dynamics simulation.
Dr. Noppadon Nuntawong
National Electronics and Computer Technology Center (NECTEC), National Science and Technology Development Agency (NSTDA), Rama VI Rd., Pathumthani, Thailand
ONSPEC: A Surface-Enhanced Raman Substrate and an Application towards Tuberculosis Diagnosis Based on Multivariate Statistical Analysis and Machine Learning
Surface-enhanced Raman spectroscopy (SERS) has gained increasing attentions towards rapid analyses of biological molecules because of its capability to detect molecules at ultralow concentrations and even the single molecule level. SERS is a technique that has been widely using in bio-application with its advantage in distinguishing a molecule by its unique vibration and enhancing the signal by a suitable nanostructured substrate. Our group have developed a commercialized SERS substrate, namely Onspec, which comprises of a uniquely fabricated nanostructure on a silicon substrate. We have experimented our Onspec substrate in many areas of applications including forensic, agriculture, environment, and medical diagnosis. SERS data analysis was typically achieved via investigating spectral pattern and changes in peak intensity, band position, the full width at half maximum (FWHM) of a band, etc. Therefore, limitations happen in the case, especially in biomolecular detections where the features are too complicate to be effectively examined by a conventional analysis technique, which normally rely on the expertise of an expert. Multivariate statistical analysis and machine learning (ML), which can be seen as a part of artificial intelligence (AI), has recently been a hot topic in many areas including spectroscopic and sensor technologies to access the prediction power from the learning of big data and can be a powerful solution to the aforementioned issue. In this work, we report the performance of the Onspec combined with multivariate statistical analysis and ML, and potential of the method in tuberculosis infection diagnosis.
Dr. Orathai Sawatdichaikul
Department of Nutrition and Health, Institute of Food Research and Product Development, Kasetsart University, 50 Ngamwongwan Rd, Chatuchak Bangkok 10900 Thailand
Protein Bioinformatics: The Applications in Food, Agrifood, and Food Fishery Research
In food, agrifood, and food fishery research, Bioinformatics is involved to elucidate the molecular mechanism of target organisms. Generally, the research studies are starting from gene characterization, clone, to DNA or mRNA sequencing. This information will be stored in the cDNA library. As the limitation of time and expense in research projects, only a few chances that protein expression study is able to perform. Interestingly, Protein Bioinformatics including protein sequence analysis, protein modeling, and protein structural analyses, is not only accelerating the study of protein structures from genes of interest, from genetic data but also facilitating deep insight into predicting functional properties of these predicted proteins. There are several successful cases to construct the three-dimensional protein structure from protein modeling tools to elucidate the behavior of gene expression, including A) the study of the myostatin (MSTN) gene from walking catfish (Clarias macrocephalus, Günther 1864) and African catfish (Clarias gariepinus), B) fragrance gene, (Badh2) from aromatic coconut (Cocos nucifera), C) cathepsin B from the giant river prawn Macrobrachium rosenbergii (MrCTSB). Moreover, bioinformatics plays a crucial role in protein design. Novel chimeric multiepitope vaccine for Streptococcosis disease in Nile Tilapia (Oreochromis niloticus Linn.) has been constructed starting from the knowledge of protein structure and protein bioinformatics, to design the core structure and the exposal active interface. The designed vaccines have been verified, giving good immune response after challenging fish with S. agalactiae. Bioinformatics could represent an important step toward removing the research barriers and is a genuine prospect for the future.
Dr. Bodee Nutho
Department of Pharmacology, Faculty of Science, Mahidol University, Bangkok 10400, Thailand
In Vitro and In Silico Investigations of C-12 Dithiocarbamate Andrographolide Analogues as Novel Sars-Cov-2 Main Protease Inhibitors
In parallel to vaccine development, the discovery of antiviral agent is another frontier research in the fight against coronavirus disease 2019 (COVID-19) caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). In particular, a modified compound derived from natural products is of potential interest for drug discovery process. An attractive drug target for COVID-19 treatment is the viral main protease (Mpro), which processes viral polyprotein during infection. In the present study, a series of twenty-one 12-dithiocarbamate-14-deoxyandrographolide analogues was explored for their anti-SARS-CoV-2 Mpro potential based on in vitro and in silico investigations. We firstly screened the inhibitory activity of these compounds toward SARS-CoV-2 Mpro using in vitro enzyme inhibition assay. The results showed that there were four promising compounds including 3k, 3l, 3m and 3t that exhibited the inhibitory activity against SARS-CoV-2 Mpro with >50% inhibition at 10 μM. Next, molecular docking was performed to determine the binding mode of each focused compound in the enzyme active site, in which the most probable complexes were selected and subjected to subsequent molecular dynamics (MD) simulations. The MD results revealed that all studied complexes were stable over the course of simulation, and most of the compounds could specifically bind to the SARS-CoV-2 Mpro active site, especially at S1, S2 and S4 subsites. The hot-spot residues involved in ligand binding were T25, H41, C44, S46, M49, C145, H163, M165, E166, L167, D187, R188, Q189 and T190. In addition, the van der Waals interactions yielded the main energy contribution stabilizing all studied compounds. Therefore, our findings from both combined experimental and computational approaches could lead to further optimizations of more potential andrographolide analogues toward SARS-CoV-2 Mpro.
Session Artificial Intelligence for Science and Engineering
Assoc. Prof. Sarana Nutanong
Vidyasirimedhi Institute of Science and Technology, Rayong, Thailand
Model-Centric vs. Data-Centric Machine Learning
Machine learning (ML) research has been model-centric for decades. That is, we consider ML modeling to be the heavy-lifting part requiring deep expertise to get it right. And the data preparation part is a tedious yet trivial job to complete as part of the pipeline. Although ensuring the data quality is understood to be a vital part of the pipeline, researchers often consider this part static. Once the data has been collected and prepared, we dedicate most of our attention to creating an accurate model. Data-centric AI challenges this concept in two different ways. First, it argues that many applications do not require a new model; the current off-the-self baseline suits the purpose. Second, it shows that effort spent on improving the data quality is more rewarding than improving the ML technique. This talk presents research and development of data-centric machine learning and comprises the following topics. First, I will outline recent works in Data-Centric ML worldwide. Second, I will talk about the industrial and academic research we are conducting at VISTEC and how we plan to bring the benefits of data-centric machine learning to our society. Third, I will propose how model-centric and data-centric efforts can synergize. Finally, this talk will conclude with implications for businesses wishing to apply ML to solve their domain-specific problems.
Dr. Akkapon Wongkoblap
DIGITECH, Suranaree University of Technology, Nakhon Ratchasima, Thailand
Identifying Depression Markers in Social Media Content
Mental health problems are widely recognized as a major public health challenge worldwide. In Thailand, several people are increasingly experiencing mental illness. Despite the increasing number of sufferers, it is still difficult for them to gain access to treatments and services. This highlights the need for effective approaches for detecting mental health disorders in the population and providing health services. Social media data is a promising source of information where people publish rich personal information that can be mined to extract valuable psychological markers. Machine learning is a potential technique for developing a predictive model capable of detecting users with depression from their social media posts and identify textual content associated with self-disclosure of their mental health disorders on social media platforms. The predictive model offers us advantages in several ways. Early detection can help plan strategic frameworks for public health to improve nation’s health, e.g., providing more mental health services in the area of the large number of sufferers. We can also develop an intervention tool to detect users with mental health problems and to provide health information or mental health services. This would stop the increasing number of mental health problems.
Dr. Naruemon Pratanwanich
Department of Mathematics and Computer Science, Faculty of Science, Chulalongkorn University, Thailand
XPORE: A New Bioinformatics Tool for RNA Modification Identification
RNA modifications such as N6-methyladenosine (m6A) have been found to contribute to molecular functions of RNAs and have been implicated in developmental process, cell-fate determination, and cancer. The emerging roles of RNA modifications in human cancer suggest that uncovering the importance of epitranscriptomics will provide powerful implications in precision oncology. However, identification of differences in RNA modifications has been challenging. Here, I will present our computational method, XPORE, to identify differential RNA modifications from direct RNA sequencing data. Based solely on the raw current intensity profiles, we extend a standard two-component Gaussian mixture model to accommodate multi-sample comparisons. For each single site, the model learns two distributions, the signal properties that are shared across samples, while allowing the probability associated with each distribution to be inferred specifically for each sample. Having incorporated prior knowledge into the model, we are able to determine the signal distributions of the modified kmers and quantitatively estimate the modification rates accordingly. XPORE was evaluated on transcriptome-wide m6A profiling data with and without replicates for prioritising differentially modified sites. In addition, the application of xPore on direct RNA-Sequencing data from 6 human cell lines revealed the landscape of RNA modifications i.e. differentially modified sites across all cell lines with cell type-specific modification rates. With xPore, I will demonstrate that RNA modifications can be quantitatively identified from direct RNA-sequencing data with high accuracy, opening many new opportunities for large scale applications in precision medicine.
Dr. Akara Supratak
Faculty of Information and Communication Technology (ICT), Mahidol University, Thailand
Tinysleepnet: an Efficient Deep Learning Model For Sleep Stage Scoring Based on Raw Single-Channel Eeg
Deep learning has become popular for automatic sleep stage scoring due to its capability to extract useful features from raw signals. Most of the existing models, however, have been overengineered to consist of many layers or have introduced additional steps in the processing pipeline, such as converting signals to spectrogram-based images. They require to be trained on a large dataset to prevent the overfitting problem (but most of the sleep datasets contain a limited amount of class-imbalanced data) and are difficult to be applied (as there are many hyperparameters to be configured in the pipeline). In this paper, we propose an efficient deep learning model, named TinySleepNet, and a novel technique to effectively train the model end-to-end for automatic sleep stage scoring based on raw single-channel EEG. Our model consists of a less number of model parameters to be trained compared to the existing ones, requiring a less amount of training data and computational resources. Our training technique incorporates data augmentation that can make our model be more robust the shift along the time axis, and can prevent the model from remembering the sequence of sleep stages. We evaluated our model on seven public sleep datasets that have different characteristics in terms of scoring criteria and recording channels and environments. The results show that, with the same model architecture and the training parameters, our method achieves a similar (or better) performance compared to the state-of-the-art methods on all datasets. This demonstrates that our method can generalize well to the largest number of different datasets.
Dr. Pattharaporn Thongnim
Burapha University, Chanthaburi Campus, 57 Moo.1 Chon Pratan Road, Kamong Sub-district, Tha Mai District, Chanthaburi Province 22170, Thailand
Applying a Mixture Model of the Gaussian Process to Agricultural Data
In the field of machine learning, it is very essential to figure out how to train the parameters of statistical models to be able to describe a given set of data. The application of Gaussian process models is increasing in the fields of statistics, engineering, and other related fields as a result of their practical utility in the real world and their interesting analytical characteristics. When the model is integrated with a non-linear model of agricultural data, the complicated time series patterns may be difficult to forecast due to heterogeneous data sets. Therefore, it is applied to the Gaussian Process Mixture (GPM) model and a learning algorithm that uses the hidden variables posterior iterative learning technique to improve forecasting performance for agricultural data. In this procedure, the expectation-maximization (EM) learning method is used to determine the most effective way to group agricultural data sets. The findings of the GPM model show that it is capable of making improved predictions with a wide range of test data, demonstrating the utility of our approach. The results also demonstrate the effectiveness and application of the proposed methodology when applied to agricultural data sets.
Department of Software Engineering, Chiang Mai University, Thailand
Research Division, Faculty of Medicine Siriraj Hospital, Mahidol University, Thailand
Multi-Scale Convolution and Multi-Resolution Processing in Sequence Modeling
Information within sequences is not necessarily encoded in fixed-length patterns. Functionally and structurally relevant motifs in protein sequences often have variable lengths and various important features of wave signals may exhibit over different durations. As a result, paralleled convolutional layers with varying kernel sizes or multi-scale convolution neural networks (CNN) have been implemented to capture these patterns of unfixed lengths across multiple fields of science and engineering. In a similar fashion, local data processing with a fixed input window such as the Short-time Fourier transform (STFT) can reveal information at different resolutions and detect the features of interest when applied at various window sizes in parallel. I will first review previous works across multiple disciplines that apply these techniques. Then I will provide an example of our project on music separation that uses multi-resolution STFT and investigates the effects of multi-scale CNN when applied to amplitude-frequency and/or phase information.
Session Mathematics and Statistics
Prof. Rabian Wangkeeree
Department of Mathematics, Faculty of Science, Naresuan University, Phitsanulok, Thailand
The Combination Algorithm of Smooth Generalized Pinball SVM and Vaes for Pneumonia Infected Patients Image Recognition
Pneumonia is the most common disease caused by various microbial species such as bacteria, viruses, and fungi that inflame the air sacs in one or both lungs. In this paper, we present a method based on combining a smooth generalized pinball support vector machine (SVM) and variational autoencoders (VAEs) in chest X-ray (CXR) images. For a smooth generalized pinball support vector machine, we try to solve the primal quadratic programming problems of SVM by transforming individuals into smooth unconstrained minimization problems. It would be of great interest to propose an efficient regularized stochastic BFGS algorithm-based SVM using the smooth generalized pinball loss function (termed as RES-SGSVM). More specifically, we have introduced verified theorems that are related to our approaches, and the theoretical convergence of our approaches has been proposed. To test the model, numerical comparisons between parameter values of several iteration objective functions are also provided. The results obtained on several benchmark datasets show that our method outperforms existing classifiers in terms of accuracy and Matthews correlation coefficient (MCC).
Prof. Juan Matinez Moreno
University of Jaén, Spain
Prof. Suthep Suantai
Chiang Mai University, Thailand
Assoc. Prof. Nirattaya Khamsemanan
Thammasat University , Thailand
Gait is a locomotion of a human. In layman’s terms, it is a way a person walks. Unlike other biometric properties of a person, gait can be collected from a far and non-intrusively. Moreover, it is hard for a person to disguise his or her gait for a long period of time. Gait recognition is a technique that use human’s gaits to identify a person. In this talk, we are going to explore gait recognition techniques using various machine learning techniques, as well as age estimation using gaits and gender recognition using gaits.
Assoc. Prof. Kittikorn Nakprasit
Department of Mathematics, Faculty of Science, Khon Kaen University
Colorings in Graphs: Proper Colorings and Some Generalizations
Graph coloring is one of the oldest and widest studied topics in graph theory. Its popularity due to its history, relations to other graph theoretical topics, and real word applications such as resource management and time scheduling. In this talk, we survey some types of coloring from a classical coloring, that is a proper coloring, to some of its generalizations namely relaxed coloring, list coloring, and DP-coloring. Furthermore, we also explore some recent researches including certain conceptual combinations of aforementioned types of coloring.
Asst. Prof. Wajaree Weera
Department of Mathematics, Faculty of Science, Khon Kaen University, Khon Kaen 40002, Thailand
Novel Stability Conditions for Genetic Regulatory Networks with Interval Time-Varying Delays
In this work, the stability analysis problem of the genetic regulatory networks (GRNs) with interval time-varying delays is presented. By constructing the Lyapunov–Krasovskii functional dependent on the composite function variables of the state estimate system, a new technique lemma is provided and applied to the stability analysis. Moreover, by adding two quadratic terms, which double integral terms of a positive quadratic term and the other one is a quadratic term which does not need to meet positive definite and would relax the stability conditions. The sufficient condition of state estimator design is obtained in terms of linear matrix inequalities or computing an induced norm of a constant matrix, which can be easily realized by usual software tools. Finally, a simulation example illustrates the rationality and effectiveness of the proposed theoretical results. The obtained results are essentially new and complement previously known results.
Dr. Abubakar Adamu
Mathematics Institute, African University of Science and Technology, Abuja 900107, Nigeria
Relaxed Modified Tseng Algorithm for Solving Variation Inclusion Problems in Real Banach Spaces with Applications
This talk will be on relaxed and relaxed inertial Tseng-type algorithms we introduced for approximating zeros of sum of two monotone operators whose zeros are fixed points or J-fixed points of some nonexpansive-type mappings. We will give a sketch of our proof that the iterates of our proposed algorithms converge strongly in the setting of real Banach spaces that are uniformly smooth and 2-uniformly convex. Furthermore, applications of our results to the concept of J-fixed point, convex minimization, image restoration, and signal recovery problems will be presented. In addition, some interesting numerical implementations of our proposed methods in solving image recovery and compressed sensing problems will be presented. Finally, we will present the performance of our proposed methods in comparison with that of some existing methods in the literature.
Session High Performance Computing
Asst. Prof. Paskorn Champrasert
Information Technology Service Center, Chiang Mai University
ERAWAN HPC: Design and Development of High-Performance Computing for Academic Research at Chiang Mai University
Cyberinfrastructure and computing resources, including the CPU, memory, data storage, and GPU, become the keys for current academic research in universities. In 2019, Chiang Mai University focused on students’ learning and working environments by providing a thousand thin clients with a virtual desktop infrastructure (VDI) system. However, with the high demand for computing resources for research, the VDI applications are redesigned to support the researchers’ requirements. The computing resource requirement assessment has been processed. The online assessment consists of 1) research information, 2) computing resource requirement, and 3) software requirements. The assessment results of 16 researchers cover various research fields (e.g., climate changes, PM2.5, chemistry model, bioinformatics, health, AI, and machine learning). In the first phase, the researchers prefer remote virtual machines over the job submission system. The average computational time is about an hour per job. This presentation contributes to the assessment results and the design of the high-performance computing system for research and innovation development at Chaing Mai University.
Dr. Thanaphon Tangchoopong
Department of Computer Science, College of Computing, Khon Kaen University
Introducing Computational Science for the First-Year Computer Science Students
Computational Science is a foundational subject that needs proficiency in mathematical modeling, numerical analysis, software engineering, high-performance computing, and statistics, in addition to a thorough understanding of the specialized application area under study. Hence, this leads to interdisciplinary among mathematicians, statisticians, application scientists, and computer scientists. As a result, in terms of computer scientists, it is challenging to design the course for introducing the freshmen in Computer Science to the subject. In this talk, the speaker would like to share the experience in educating computational science in terms of concepts necessary for the students, teaching paradigms, and how the assessment should be.