Memorial Lectures

The following commemorative lectures will be held on Monday, October 28, following the opening ceremony at 17:30. 

Considering drug withdrawals due to unexpected toxicity and the success of drug repositioning based on newly discovered effects, drugs have numerous unknown effects. To uncover these unrecognized aspects, we promote a simple data-driven approach with low-biased numerization and pattern recognition, enabling comprehensive drug profiling akin to a radar chart. How can we exclude bias stemming from our recognition capacity in describing drug properties? One method is employing comprehensiveness, like omics analysis. By measuring all variables in a layer, an object of interest can be described without bias regarding that layer at least. For instance, a drug's effect can be considered biological responses, often represented as omics data from a treated specimen. Another approach involves using data close to our sensory systems, such as images, which can be obtained independently of human recognition, thus reducing bias stemming from recognition capacity. However, life science information in these data often shows invariance, like rotation invariance in images. Neural networks are useful for extracting latent representations to address such issues. We are focusing on the numerization and pattern recognition of various life science data related to drugs such as transcriptome data, toxicopathological data, and chemical structures. Here, we introduce recent advances in our work.

Recent developments in biotechnology have contributed to the increase in the amounts of high-throughput data in the genome, transcriptome, proteome, interactome, phenome and diseasome. These biomedical big data can be useful resources for drug development processes. Bioinformatics technologies are expected to play key roles in the big data analysis. In this study, we developed novel bioinformatics methods to predict therapeutic targets of diseases, to search for drug candidate molecules, and to design new chemical structures of drug candidate molecules, by integrating various biomedical data on compounds (e.g., chemical structures, clinical phenotypes, gene expression patterns, target molecules) and diseases (e.g., disease-causing genes, environmental factors, and clinical information). A unique feature of our data-driven approach is that it clarifies all target proteins of each drug including off-targets, estimates the mechanisms of action at the pathway level, and generates molecular structures of drug candidates by deep learning. In my talk at the conference, we will show some of the applications to therapeutic target identification, large-scale compound screening, combination therapy, drug molecular structure design for a variety of diseases, and target-based prediction of toxicity and side effects.