Review
Chemistry, Multidisciplinary
Bing Huang, O. Anatole von Lilienfeld
Summary: Chemical compound space (CCS) is vast and exploring it using modern machine learning techniques based on quantum mechanics principles can improve computational efficiency while maintaining predictive power. These methods have potential applications in discovering novel molecules or materials with desirable properties.
Article
Computer Science, Software Engineering
Maria Virginia Sabando, Pavol Ulbrich, Matias Selzer, Jan Byska, Jan Mican, Ignacio Ponzoni, Axel J. Soto, Maria Lujan Ganuza, Barbora Kozlikova
Summary: In the modern drug discovery process, medicinal chemists use computational tools like dimensionality reduction and classification to analyze large ensembles of candidate molecules. To address the complexity of such data, ChemVA is introduced as an interactive application to visually explore large molecular ensembles and their features, allowing users to efficiently select candidate compounds through multiple coordinated views. This system enables effective visual inspection and comparison of different high-dimensional molecular representations while providing information on the certainty behind these representations.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2021)
Article
Computer Science, Information Systems
Mostafa M. Moussa, Usman Tariq, Fares Al-Shargie, Hasan Al-Nashash
Summary: This study developed an experimental protocol to differentiate genuine and fake smile expressions using EEG signals and CNN models. The results indicate that CNN models can effectively detect smiles from EEG signals.
Article
Biology
Dixin Zhou, Fei Liu, Yiwen Zheng, Liangjian Hu, Tao Huang, Yu S. Huang
Summary: This paper presents a deep learning-based virtual screening neural network model called Deffini, which performs well on benchmark datasets but has poor performance on the MUV dataset. The author discovers that family-specific models outperform pan-family models and explores the limits of predictive power using a newly constructed protein kinase dataset called Kernie.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Biochemical Research Methods
Xiao-Chen Zhang, Jia-Cai Yi, Guo-Ping Yang, Cheng-Kun Wu, Ting-Jun Hou, Dong-Sheng Cao
Summary: This paper presents a deep neural network model called ABC-Net, which can directly predict graph structures. By using the divide-and-conquer principle, atoms or bonds are modeled as single points in the center, and a fully convolutional neural network is leveraged to identify and predict relevant properties, enabling the recovery of molecular structures. Experimental results demonstrate significant improvement in recognition performance with this method.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Chemistry, Medicinal
Jack Scantlebury, Lucy Vost, Anna Carbery, Thomas E. Hadfield, Oliver M. Turnbull, Nathan Brown, Vijil Chenthamarakshan, Payel Das, Harold Grosjean, Frank von Delft, Charlotte M. Deane
Summary: In recent years, multiple machine learning-based scoring functions have been developed to predict the binding of small molecules to proteins. These scoring functions aim to approximate the distribution that takes two molecules as input and outputs the energy of their interaction. However, many scoring functions rely on data set biases rather than understanding the physics of binding. To test the learning ability of machine learning-based scoring functions, input attribution can be applied to identify important binding interactions. A machine learning-based scoring function was built, which achieved comparable performance to other methods on benchmark tests. Attribution was then used to extract important binding pharmacophores and improve docking scores compared to traditional approaches.
JOURNAL OF CHEMICAL INFORMATION AND MODELING
(2023)
Review
Chemistry, Physical
Shilpa Shilpa, Gargee Kashyap, Raghavan B. Sunoj
Summary: The burgeoning developments in machine learning have found notable applications in chemistry, particularly in the prediction of molecular properties and chemical reactions. This review highlights the recent advancements in ML implementations, ranging from ensemble-based models to graph neural networks. Accurate predictions in molecular property prediction, using methods such as D-MPNN, MolCLR, SMILES-BERT, and MolBERT, offer promising prospects in molecular design and drug discovery. Challenges in dealing with reaction data sets are discussed, alongside an optimistic outlook on the benefits of ML-driven workflows for various chemistry tasks.
JOURNAL OF PHYSICAL CHEMISTRY A
(2023)
Article
Cardiac & Cardiovascular Systems
Michal Cohen-Shelly, Zachi Attia, Paul A. Friedman, Saki Ito, Benjamin A. Essayagh, Wei-Yin Ko, Dennis H. Murphree, Hector Michelena, Maurice Enriquez-Sarano, Rickey E. Carter, Patrick W. Johnson, Peter A. Noseworthy, Francisco Lopez-Jimenez, Jae K. Oh
Summary: The study aimed to develop an artificial intelligence-enabled electrocardiogram (AI-ECG) to identify patients with moderate to severe aortic stenosis (AS). The AI-ECG showed promising results in detecting AS, especially in a community screening setting.
EUROPEAN HEART JOURNAL
(2021)
Article
Cardiac & Cardiovascular Systems
Michal Cohen-Shelly, Zachi Attia, Paul A. Friedman, Saki Ito, Benjamin A. Essayagh, Wei-Yin Ko, Dennis H. Murphree, Hector Michelena, Maurice Enriquez-Sarano, Rickey E. Carter, Patrick W. Johnson, Peter A. Noseworthy, Francisco Lopez-Jimenez, Jae K. Oh
Summary: The study showed that AI-ECG has a certain degree of accuracy and reliability in identifying patients with moderate to severe AS, especially showing differences among patients of different ages and genders. For patients with false-positive AI-ECGs, the risk of developing moderate or severe AS within the next 15 years is twice that of true negative AI-ECGs.
EUROPEAN HEART JOURNAL
(2021)
Article
Computer Science, Information Systems
R. Ani, O. S. Deepa
Summary: This study proposes a deep learning based virtual screening model for the early discovery of drug compounds for Hemochromatosis. The model outperformed other models in accuracy and F-score and identified a small set of biologically active compounds. In-vitro studies are recommended for these compounds.
Article
Multidisciplinary Sciences
Yukihiro Nomura, Masato Hoshiyama, Shinsuke Akita, Hiroki Naganishi, Satoki Zenbutsu, Ayumu Matsuoka, Takashi Ohnishi, Hideaki Haneishi, Nobuyuki Mitsukawa
Summary: This study developed a computer-aided diagnosis software using deep learning for screening lower extremity lymphedema (LEL) in pelvic CT images. By using fat-enhanced images and the ResNet-34 model, high screening accuracy was achieved.
SCIENTIFIC REPORTS
(2023)
Article
Computer Science, Artificial Intelligence
Fouaz Berrhail, Hacene Belhadef, Mohammed Haddad
Summary: LBVS plays a crucial role in the early stage of drug discovery, solving time and cost issues associated with traditional methods. The proposed method based on Deep Convolutional Neural Network (DCNN) demonstrates superior performance compared to conventional methods through new learning representation for chemical compounds.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Biochemistry & Molecular Biology
David Bacelar Costa Junior, Janay Stefany Carneiro Araujo, Larissa de Mattos Oliveira, Flavio Simas Moreira Neri, Paulo Otavio Lourenco Moreira, Alex Gutterres Taranto, Amanda Luisa Fonseca, Fernando de Pilla Varotti, Franco Henrique Andrade Leite
Summary: Malaria, caused by Plasmodium spp., remains a significant public health issue with drug resistance limiting chemotherapy. By exploring essential therapeutic targets of the parasite, a potential new antimalarial molecule with activity comparable to artemether was identified through virtual screening. This molecule, previously used in traditional Chinese medicine, may offer a repurposing opportunity to expedite drug development. Molecular dynamics studies confirmed the stability of molecular interactions, paving the way for further research on new derivatives for in vitro and in vivo evaluation.
JOURNAL OF BIOMOLECULAR STRUCTURE & DYNAMICS
(2021)
Article
Engineering, Chemical
Shengli Jiang, Victor M. Zavala
Summary: This article reviews the mathematical foundations of convolutional neural networks (CNNs) and their connections with statistics, signal processing, linear algebra, differential equations, and optimization techniques. CNNs are powerful models that highlight features from grid data to make predictions, and they can be applied to a wide range of applications beyond image and video processing. The operators used in CNNs are learned through optimization techniques to best map input data to output data, allowing for flexibility in representing grid data with multiple channels.
Article
Chemistry, Medicinal
Fahad Hassan Shah, Song Ja Kim
Summary: The study identified three natural compounds (betanine, hesperetin and ovatodiolide) that can effectively inhibit FGL2 protein and potentially serve as candidate drugs for treating FGL2-mutated glioblastomas.
FUTURE MEDICINAL CHEMISTRY
(2021)
Article
Biochemistry & Molecular Biology
Akito Taneda, Kengo Sato
Summary: The programmability of RNA-RNA interactions has been successfully utilized to design RNA devices that regulate gene expression, but designing structured RNA sequences that meet multiple criteria has become a complex problem. Despite the lack of a web service for multi-objective design of RNA switches utilizing RNA-RNA interactions, a web server based on the MODENA algorithm was developed to design two interacting RNAs in silico.
INTERNATIONAL JOURNAL OF MOLECULAR SCIENCES
(2021)
Article
Multidisciplinary Sciences
Kengo Sato, Manato Akiyama, Yasubumi Sakakibara
Summary: Combining thermodynamic information with deep learning can improve the robustness of RNA secondary structure prediction compared to existing algorithms.
NATURE COMMUNICATIONS
(2021)
Article
Multidisciplinary Sciences
Yoshihiro Miyazaki, Tatsuya Oda, Yuki Inagaki, Hiroko Kushige, Yutaka Saito, Nobuhito Mori, Yuzo Takayama, Yutaro Kumagai, Toutai Mitsuyama, Yasuyuki S. Kida
Summary: Cancer-associated fibroblasts (CAFs) play important roles in tumor progression and drug resistance in pancreatic ductal adenocarcinoma (PDAC), but existing mouse models have limitations in reproducing the characteristics of clinical CAFs. Researchers have developed a new human cell-derived stroma-rich CDX model, which successfully recapitulates the clinical features of pancreatic cancer by co-transplanting human adipose-derived mesenchymal stem cells and a PDAC cell line into mice.
SCIENTIFIC REPORTS
(2021)
Article
Multidisciplinary Sciences
Shin Irumagawa, Kaito Kobayashi, Yutaka Saito, Takeshi Miyata, Mitsuo Umetsu, Tomoshi Kameda, Ryoichi Arai
Summary: This study successfully predicted and experimentally confirmed new mutations that improve protein stability through in silico saturation mutagenesis and molecular dynamics simulation. The double mutant N22A/H86K showed significant improvement, and these thermostable mutants have potential significance for constructing supramolecular protein complexes.
SCIENTIFIC REPORTS
(2021)
Article
Biochemical Research Methods
Hideki Yamaguchi, Yutaka Saito
Summary: Accurate variant effect prediction plays a significant role in protein engineering. Recent machine learning approaches focus on representation learning to generate feature vectors from unlabeled sequences. This article proposes DA-aware evolutionary fine-tuning protocols for Transformer-based variant effect prediction, achieving better performances than previous methods and incorporating structural information without direct supervision.
BRIEFINGS IN BIOINFORMATICS
(2021)
Article
Biotechnology & Applied Microbiology
Naoyuki Tajima, Toshitaka Kumagai, Yutaka Saito, Tomoshi Kameda
Summary: The relationship between translation efficiency and sequence features varies across organisms, reflecting their taxonomy. The codon adaptation index shows high correlation in all analyzed organisms.
Article
Mathematical & Computational Biology
Godai Suzuki, Yutaka Saito, Motoaki Seki, Daniel Evans-Yamamoto, Mikiko Negishi, Kentaro Kakoi, Hiroki Kawai, Christian R. Landry, Nozomu Yachie, Toutai Mitsuyama
Summary: Morphological profiling, combining optical microscopes and machine vision technologies, has been successfully applied in high-throughput phenotyping. The study demonstrates the potential to discriminate single-gene mutant cells from wild-type cells based on bright-field images. Machine learning was used to construct a model that successfully identified mutant cells.
NPJ SYSTEMS BIOLOGY AND APPLICATIONS
(2021)
Article
Chemistry, Physical
Yutaka Saito, Misaki Oikawa, Takumi Sato, Hikaru Nakazawa, Tomoyuki Ito, Tomoshi Kameda, Koji Tsuda, Mitsuo Umetsu
Summary: The study shows that machine learning is a useful tool in designing proteins with desired functions in protein engineering. Depending on the presence or absence of highly positive variants in the training data, machine learning-guided directed evolution can lead to improved variants in different regions of sequence space.
Article
Biology
Shunya Kashiwagi, Kengo Sato, Yasubumi Sakakibara
Summary: Protein-RNA interactions are crucial for biological processes, and various computational methods have been developed to predict these interactions. However, accurately predicting residue-base contacts in PRIs remains a challenge. The proposed method using only sequence and predicted structural information shows promising results, comparable to methods based on known binding data.
Article
Biochemical Research Methods
Soichiro Nishiyama, Kengo Sato, Ryutaro Tao
Summary: This study presents a novel approach for selecting informative markers based on binary integer programming. By combining with targeted SNP genotyping, this method allows for flexible analysis and has practical applications in large-scale problems in breeding and ecological research.
BMC BIOINFORMATICS
(2022)
Article
Medicine, Research & Experimental
Tomoyuki Ito, Thuy Duong Nguyen, Yutaka Saito, Yoichi Kurumida, Hikaru Nakazawa, Sakiya Kawada, Hafumi Nishi, Koji Tsuda, Tomoshi Kameda, Mitsuo Umetsu
Summary: The aim of this study was to design an improved library based on information from a weakly enriched library, as bias during biopanning often leads to the enrichment of undesired variants. Deep sequencing of previous biopanning results revealed that weak enrichment was partially due to biases during phage infection and amplification steps. Machine learning analysis of the deep sequencing data identified distinct sequence patterns, which were used to design phage libraries. Four improved variants with specific target affinity were identified using biopanning.
Article
Biochemical Research Methods
Yuki Ogawa, Yutaka Saito, Hideki Yamaguchi, Yohei Katsuyama, Yasuo Ohnishi
Summary: Enzyme engineering using machine learning has made significant progress in recent years. This study explores the application of biosensor-based enzyme engineering method in machine learning. By evaluating the productivity of XylM variants using a fluorescence intensity-based biosensor, training data for machine learning was obtained and a XylM variant with 15 times higher productivity than wild-type XylM was successfully obtained. These findings demonstrate the quantitative and high-throughput capability of biosensors in indirect enzyme activity evaluation, expanding the versatility of machine learning in enzyme engineering.
ACS SYNTHETIC BIOLOGY
(2023)
Article
Genetics & Heredity
Manato Akiyama, Yasubumi Sakakibara, Kengo Sato
Summary: This study proposes a new algorithm for directly inferring base-pairing probabilities of RNA secondary structures using neural networks, independent of their architecture. The algorithm outperforms existing methods in prediction accuracy, as demonstrated by benchmarks with and without pseudoknots.
Review
Biochemical Research Methods
Kengo Sato, Michiaki Hamada
Summary: Computational analysis of RNA sequences plays a crucial role in RNA biology. In recent years, the incorporation of artificial intelligence and machine learning techniques into RNA sequence analysis has gained significant traction. Machine learning-based approaches have shown remarkable advancements, enhancing the precision of sequence analysis related to RNA secondary structures. Furthermore, artificial intelligence and machine learning innovations are also applied in the analysis of RNA-small molecule interactions, RNA drug discovery, and the design of RNA aptamers.
BRIEFINGS IN BIOINFORMATICS
(2023)
Article
Chemistry, Multidisciplinary
Jumpei Maki, Asami Oshimura, Chihiro Tsukano, Ryo C. Yanagita, Yutaka Saito, Yasubumi Sakakibara, Kazuhiro Irie
Summary: Protein kinase C (PKC) family is a potential target for treating cancer, Alzheimer's disease, and HIV infection. By screening compounds and designing analogues, we discovered a PKC ligand with remarkable isozyme selectivity.
CHEMICAL COMMUNICATIONS
(2022)