Program

Oct.29 Tue 10:10〜10:40 Tower Hall Funabori 5F Main Hall

In the application of artificial intelligence (AI) and machine learning in biological sciences, and drug discovery research in particular, ensuring the quality and quantity of the training data is considered to be one of the most important factors for building successful predictive models. While the development of AI-driven drug discovery platforms and their applications to individual projects are being actively pursued, no single solution can address the issue of data quality and quantity. In my talk, I will discuss this issue by describing our recent research as case studies, attempting at 1) establishing novel data integration and curation strategies, and 2) colleting new data using own experiments. Specific examples include: the prediction of nano particle safety, which utilized literature data mining and in-house experimental data (Martin et al., ACS Nano, 2023), stability of antisense oligonucleotides including artificial nucleic acids (Kuroda et al., Mol. Ther. Nucleic Acids, 2024), another example of the use of in-house data obtained with a new experimental technique, and the contribution ratio of a metabolic enzyme to in vivo clearance (Watanabe et al., Mol. Pharmaceutics, 2022), with intensive expert-curation of public data.

Oct.29 Tue 10:40〜12:00 Tower Hall Funabori 5F Main Hall

Yayoi Natsume (National Institutes of Biomedical Innovation, Health and Nutrition)

In the pharmaceutical industry, the number of high-evidence drug targets is decreasing, making drug discovery more challenging. Meanwhile, the external environment in drug discovery is rapidly changing due to advancements in ML/AI technologies, which significantly improve data analysis processing speeds. Leveraging these cutting-edge digital and IT technologies is expected to substantially streamline parts of the drug discovery process while significantly improving the success rate of drug development. We are actively utilizing AI technologies to drive transformations in the drug discovery process, including the exploration of drug candidate molecules, evaluation of efficacy and safety through image analysis, literature search and drug target identification using natural language processing, and development of laboratory automation using robotics. Specifically, we have developed a proprietary machine learning algorithm (MALEXA®) for optimizing the selection of clinical candidate molecules for antibody drugs, which has been successfully applied to drug discovery projects. In this session, we will introduce the technological development of MALEXA®, AI application cases developed in collaboration with other companies, generative AI, growing of digital capability for our researchers, as well as our current challenges and future prospects in this field.

The evolution of deep learning AI, which began in the 2010s, is unstoppable, and recently large-scale language models, such as ChatGTP, and generative AI based on them, have made significant progress and begun to permeate our lives. Within this society-wide trend, large-scale and full-scale applications of AI, including generative AI, are becoming a reality in the fields of drug discovery and life sciences, and initial trajectory results are being obtained. This presentation will first introduce NVIDIA's overall approach to AI, followed by specific examples from the life sciences sector. In particular, AI-based drug discovery ventures, known as tech-bio in Europe and the US, and some giant pharmaceutical companies have already integrated various types of AI into their daily operations as a commonplace tool. To this end, they are developing highly efficient research and development by learning and building their own large-scale foundation models, combining AI prediction and generation with high-throughput simulations and wet experimental results in a 'lab-in-a-loop' fashion, and continuously learning and enhancing AI from these experimental data.  
To realize such advanced AI drug discovery, dozens to hundreds of GPU servers are being installed and utilized as large-scale computing platforms. Through this case study, we would like to consider the challenges for realizing AI drug discovery in Japan in the era of generative AI. 

Oct.29 Tue 14:00〜15:30 Tower Hall Funabori 5F Main Hall

Michihiro Araki(National Institutes of Biomedical Innovation, Health and Nutrition)

Epidemiological studies have shown that maintaining higher physical activity and reducing longer sedentary behavior are effective in preventing lifestyle-related diseases, heart disease, cerebrovascular disease, and the part of cancers. In line with the revision of Health Japan 21, the Ministry of Health, Labor and Welfare created guidelines on physical activity, and in January 2024 published the "Physical Activity and Exercise Guide for Health Promotion 2023." Recent physical activity research can be broadly divided into two categories. The first is research on the relationship between detailed analysis of individual physical activity (and sedentary behavior) and health outcomes, as measurement devices have evolved from questionnaires to accelerometers (pedometers) worn on the waist and accelerometers (smart watches) worn on the arm. The other is research on disease onset to interact in physical activity, lifestyle habits other than physical activity, and drug by integrating claims data and health checkup data. Both studies contain a large amount of information, and integration with data science is essential for social implementation from analysis. In this presentation, we would like to provide a deeper understanding of the outline of “Physical Activity and Exercise Guide for Health Promotion 2023” and to discuss the challenges and expectations in combining physical activity research and data science.

After disasters, lifelines such as gas and electricity are limited. Furthermore, there are health problems such as stress accumulates in disaster affected areas. On the other hand, life in space also has limited lifelines. Additionally, living in a closed space for long periods of time can be extremely stressful. 
We found that there are many similarities in disaster food and space food, where the environments are similar. When we compared the certification standards of "Japanese Disaster Food" certified by the Japanese Society for Disaster Foods and "Japanese Space Food" certified by JAXA, we found commonalities such as room temperature storage, strong packaging and hygiene management in facilities[1].  Therefore, evidence and know-how regarding food and nutrition in special environments such as after disasters are considered to be useful for improving the food environment in the space environment. Furthermore, we started training course for professionals to provide food and nutrition support in space. Currently, various activities are being carried out with the aim of staying 1,000 people to the moon by the 2040s. We would like to introduce you to the latest information. 
[1]Tsuboyama-Kasaoka N, Hamanaka K, Kikuchi Y, Nakazawa T. Similarities between Disaster Food and Space Food. J Nutr Sci Vitaminol. 2022;68(5):460-469. 

In our super-aged society, addressing lifestyle diseases and improving physical functions to achieve healthy longevity is critical. Lifestyle diseases, including obesity and diabetes, are often caused by imbalanced diet and lack of exercise, resulting from imbalances in energy intake and expenditure. Therefore, improving energy metabolism is essential. Functional food factors are gaining attention for their role in enhancing energy metabolism and physical functions. Our research highlights β-conglycinin, a soybean protein known for its various physiological functions, including an improvement of lipid metabolism. Our studies of transcriptome analysis using hepatic RNA show that β-conglycinin strongly induces the anti-obesity hormone FGF21. An improvement of lipid metabolism by β-conglycinin was diminished in FGF21-deficient mice, suggesting that FGF21 mediates the anti-obesity effects of β-conglycinin. Additionally, recent studies reveal that gut microbiota plays a crucial role in host metabolic homeostasis. From a screening by a reporter assay, we found that γHYD and γKetoD, gut bacterial metabolites of polyunsaturated fatty acid γ-linolenic acid potently activate the nuclear fatty acid receptor PPARδ. γHYD and γKetoD directly bind to PPARδ and work as ligands. We also found these metabolites improve lipid metabolism in human intestinal organoids. An exploratory food research using food factor libraries and the potential of AI will be also discussed.

Oct.30 Wed 10:00〜11:30 Tower Hall Funabori 5F Main Hall

Nobuyo Tsuboyama-Kasaoka (National Institutes of Biomedical Innovation, Health and Nutrition)

Polyphenols are well known functional components derived from plants, numerous studies have reported the biological activities. On the other hand, the lower bioavailability and unclarified biological mechanisms are called polyphenol paradox. On the other hand, their low bioavailability and many unresolved aspects of their mechanisms of effect have been called the polyphenol paradox. Interestingly, polyphenol-rich foods such as coffee and apple, as well as intestinal bacterial metabolites of polyphenols with known efficacy such as γ-oryzanol and curcumin, commonly contain the C6-C3 compound 3-(4-hydroxy-3-methoxyphenyl) propionic acid (HMPA). We have hypothesized that HMPA, an intestinal bacterial metabolite of polyphenols, is one of the forms in which polyphenols express their functions, and have been studying its efficacy, mechanism of activity, and production by fermentation. In this presentation, we will report the efficacy of HMPA in humans and discuss the significance of polyphenol metabolism by intestinal bacteria.We will also introduce our bioinformatics efforts and expectations. The traditional concept of "food" was to take in nutrients and functional ingredients, but in recent years, the concept of "food" has been changing to an understanding of the complex relationships among the three elements of human, food, and intestinal bacteria. We consider that the use of bioinformatics will become important as each element has a vast amount of data and is becoming more complex.

Dairy products include many nutrients related to bone health, such as calcium, phosphorus, and protein. However, the associations between dairy products and bone biomarkers remain unclear in Japanese. In this study, we analyzed a cross-sectional data from 1063 Japanese adults who participated in the Iwaki Health Promotion Project in 2015 to explore the associations among blood bone turnover markers, osteo sono assessment index (OSI) measured by a quantitative ultrasound technique, and dairy consumption using a food frequency questionnaire. Multivariate linear regression models were developed by adjustments for age and sex. There were significant differences in OSI Z score, a procollagen type I N-terminal peptide (P1NP), the undercarboxylated osteocalcin (ucOC), P1NP to bone alkaline phosphatase ratio (P1NP/BAP) in the habitual consumption of dairy products. For the multivariate linear regression analyses with adjustments, the low-fat dairy consumption was associated with the tartrate-resistant acid phosphatase 5b (TRACP-5b), OSI T-score, and OSI Z-score (β=–0.309, 0.002, and 0.002, P=0.026, 0.136, and 0.116). The high-fat dairy consumption was associated with the parathyroid hormone concentrations (PTH), ucOC, and P1NP/BAP after adjusting age and sex (β=0.041, 0.003, and 0.002, P=0.040, 0.014, and 0.009, respectively). The total dairy consumption was associated with PTH (β=–0.036, P=0.012). For the sensitivity analyses, in the postmenopausal females, there were statistically significant results, including TRACP-5b, OSI toward the low-fat dairy consumption and ucOC and P1NP/BAP toward the high-fat dairy consumption (P<0.05). Meanwhile, there were significant differences in P1NP, P1NP/BAP, and OSI T-score for the treatment status of osteoporosis. These findings suggest that the consumption of dairy products is partially associated with bone biomarkers in Japanese adults, mainly postmenopausal women.

Metabolic syndrome (MetS) and locomotive syndrome (LS) are two major routes caused by lifestyle habits that make it difficult to maintain health. Kao Corporation has contributed to society by developing a medical device that can easily measure visceral fat area (VFA). However, research into the impact of VFA on these two syndromes remains limited. Therefore, we introduced this device to the Hirosaki University COI-NEXT “Iwaki Health Promotion Project (IHPP)” and have been building a longitudinal big data set (10 years, 2000-3000 health checkup items, a total of 10,000 participants). Regarding the onset of MetS, VFA is considered a major factor, and we have clarified that lifestyle habits such as diet, exercise, and sleep habits were associated with the VFA accumulation. Additionally, we have developed and introduced an app that allows easy to estimate of VFA using a smartphone.On the other hand, regarding the onset of LS, musculoskeletal disorders are considered major factors, and the direct relationship between MetS and LS has not been clarified. Therefore, to reveal the relationship, we conducted a cross-sectional analysis of VFA and LS using data obtained from annual health checkups between 2015 and 2019 as a part of the IHPP. As a result, we have found that VFA is cumulatively associated with early stage of LS (LS1) in relation to age. Furthermore, using Bayesian network analysis to graphically represent the casual relationship among LS-related items, we have identified a path of items within one of the onset patterns of LS1 driven by the accumulation of VFA.In this presentation, we review the history of Kao’s research on VFA and present the latest research findings from the analysis of its relationship with LS. 

Suntory Wellness Limited was established as the company responsible for the health-related businesses within the Suntory Group and provides products and services to support customer’s wellness realization. In Japan, which is facing a super-aging society, Suntory Wellness has developed mass-market health foods based on aging research that focuses on changes in the body caused by aging. In recent years, "personalized nutrition", which suggests meals suitable for each individual according to their lifestyle and gut microbiota, has been attracting attention. Suntory Wellness is also developing products and services tailored to each individual condition. In personalized nutrition research, research focusing on postprandial blood glucose level is progressing. Research groups in the UK and Israel have reported that there are individual differences in post-meal blood glucose levels even when the same meal is consumed, and that an individual's postprandial blood glucose responses to various nutritional compositions can be predicted based on gut microbiota, lifestyle and health status. Now we are trying to create a postprandial blood glucose prediction algorithm based on the data from Japanese to develop personalized products and services. In this seminar, we would like to introduce our personalized nutrition research aimed at developing products and services tailored to each individual customer.

Oct.31 Thu 10:00〜11:30 Tower Hall Funabori 5F Main Hall

Yayoi Natsume (National Institutes of Biomedical Innovation, Health and Nutrition)

The "research" of modern science began with the observation of visible phenomena and expanded its scope by using instruments to broaden its field of view. Eventually, mathematics became the eyes and hands of science, leading to the flourishing of “engineering". Today, all phenomena, including human cognition and consideration, are "computed" as simulation models on computers.
In this presentation, I will start from the hypothesis that "programming was created for research and development." I will then overview why Python has become widely used and how it has been utilized. In programming, "data structures" serve as the eyes, and "algorithms" serve as the hands. The development of programming methodologies can be viewed as the evolution of these two methods. By examining data structures and algorithms, we can identify the "models" that connect research and development with Python. By looking at the trajectory of these models, I aim to make a simple near-future prediction about how Python will continue to be used in research and development.

RDBMS is first developed in the 1970s, and even now, more than 50 years later, they are still the most commonly used method of data management in the world. Recently new RDBMS appeared, such as cloud services that offer not only self-managed and fully-managed services, but also RDBMS-based database services with enhanced functionality. In this context, open-source databases are coming major. In Japan, PostgreSQL and MySQL are well known. For reserchers, in order to efficiently extract data from the data collected under unified management, the power of RDBMS will be necessary. The speaker has been involved in the PostgreSQL over 20 years. I am very honored to have the opportunity to introduce PostgreSQL.

This presentation explains the process of gradually revealing the mysteries from the lives of the ancient Egyptians to the construction of the pyramids using state-of-the-art ICT that makes full use of image processing technology and AI. We discover how international research teams from Japan, the USA, the UK, the Czech Republic, Poland and other countries are investigating the 4,500-year-old giant structures that remain in the desert, recording their current state and reconstructing their past, including the technology used in the TV programme "Discover the World's Mysteries!” The latest results will be shared, including 3D shape reconstruction from aerial photography of pyramids by drones and image analysis using AI.

Oct.31 Thu 14:00〜15:30 Tower Hall Funabori 5F Main Hall

Masaaki Kotera (Preferred Networks, Inc.)

AI is becoming an indispensable tool for life sciences research. We survey the landscape of AI applications, including discovery, interpretation of data and getting the right medicines to the right patients right now. Shweta will outline how AI can shorten research timelines, reduce costs, and uncover insights otherwise unattainable with traditional methods using AI. Together, we envision a future where AI drives groundbreaking discoveries across life sciences.

In today's landscape, drug discovery research is characterized by high costs and lengthy timelines, with a noticeable decline in the success rate of new drug development. Amidst these challenges, there is a growing demand for the innovative application of AI technologies, particularly generative AI, to enhance and revolutionize the drug discovery and clinical development processes. The utilization of generative AI is focused on uncovering how it can significantly contribute to improving efficiency and fostering innovation within pharmaceutical research, thereby opening up new avenues for future drug development. The application of generative AI aims to drastically reduce the time and resources required in the drug discovery and clinical development phases. This includes facilitating the discovery of target molecules and potential compound candidates, refining clinical trial designs, and automating the generation of documentation necessary for regulatory approvals. Furthermore, generative AI is being leveraged to improve internal communications, share knowledge more effectively, and extract valuable insights, all of which contribute to heightened productivity and the creation of new value within organizations. A critical aspect of incorporating generative AI into pharmaceutical research is risk management. Establishing robust guidelines and governance structures is essential for ensuring the safe and effective use of generative AI technologies. These efforts are driven by the goal of maximizing the value delivered to patients and healthcare professionals through the swift introduction of innovative medicines and services. The presentation aims to provide specific examples of these initiatives, offering a detailed exploration of how generative AI is set to transform the drug discovery process. By enhancing both efficiency and innovation, generative AI holds the promise of significantly accelerating the development and delivery of innovative medicines and services, marking a significant leap forward in the pharmaceutical industry.

Natural language processing is a research field that aims to realize computers that can understand and generate human language. ChatGPT, released by OpenAI in November 2022, caused a worldwide boom with its surprisingly professional and natural responses to a wide range of questions. Since then, research and development of large language models (LLMs) have been active, with OpenAI's GPT-4 and GPT-4V, Google's Bard and Gemini, and Anthropic's Claude 3 having been released. In addition, many companies and research institutions are developing LLMs that are fluent in Japanese. Furthermore, multimodal foundation models integrating LLMs with other modalities such as images, audio, and video are also being actively developed.In this talk, I will explain the foundations of the development and use of LLMs, including the fundamental mechanisms and development (e.g., pre-training, instruction tuning, alignment, and evaluation). I will also present continual pre-training as an approach to adapt LLMs to specific languages and fields. As an example of this approach, I will introduce Swallow, LLMs being developed at the Institute of Science Tokyo (formally known as Tokyo Institute of Technology) and the National Institute of Advanced Industrial Science and Technology (AIST). Finally, I will summarize the recent advances of LLMs in the medical domain such as Medprompt and Med-Gemini.

We have witnessed remarkable progress in Large Language Models (LLMs), accelerating their integration across various sectors of society. LLMs demonstrate unprecedented common sense understanding and advanced reasoning capabilities trained on vast datasets, The performance of LLMs continues to improve through increased training data, model size, and computational power, as well as enhancements in data quality and learning techniques. These advancements have rapidly expanded LLM applications in fields requiring extensive and specialized knowledge, such as healthcare and drug discovery. For instance, in 2024, LLMs surpassed the passing score on Japan's National Medical Examination, achieving a level where they can provide basic medical question-answering. In life sciences and chemistry, LLMs now possess knowledge comparable to experts. These systems are expected to accelerate research and development in healthcare and drug discovery, and serve as support tools in clinical settings. However, applying LLMs in healthcare and drug discovery fields presents challenges, including data confidentiality, ethical considerations, ensuring transparency in model decision-making, and eliminating misinformation and biases. Legal and regulatory frameworks also need development. This presentation will provide an overview of the latest advancements in LLMs and illustrate their potential applications in healthcare and drug discovery. Additionally, I will discuss the challenges that must be addressed, and consider the responsible use of LLMs and their future prospects.