Host: The Japanese Society of Toxicology
Name : The 51st Annual Meeting of the Japanese Society of Toxicology
Date : July 03, 2024 - July 05, 2024
With the advancement of large-scale omics technologies, toxicogenomics has emerged as a key field in safety pharmacology and chemical risk assessment. Traditional statistics-based bioinformatics analysis poses challenges in its application across multidimensional toxicogenomic data, including administration time, dosage, and gene expression levels. Leveraging advancements in image-based AI models offers a promising direction for the in visio analysis of toxicogenomics data. The Percellome Project[1], initiated in 2003, has compiled an extensive database of transcriptome time series for toxicology, focusing on multi-time and dose toxicity which is a treasure trove of novel biological insights. First, we developed Dtox [2], a deep neural network-based in visio approach that uses 3D surface plots of gene expressions to identify patterns of gene upregulation and downregulation. Unlike traditional bioinformatics methods that rely on numerical values, DTox learns from image representations, enabling the classification of genes based on experts’ labels of significance. We show that this approach outperforms statistical bioinformatics methods that rely on numerical expression values. Next, building on the foundation of the Dtox model, we deep dive into advancing these in visio models to uncover time-dependent cascades of gene network activations and suppressions. We propose a novel method that employs autoencoders to learn latent representations of gene expressions from 3D plots. These representations are then used to cluster genes by their co-expression patterns over time in an unsupervised manner. We show the effectiveness of our methods designated as ToxEye for uncovering novel insights on select datasets of the Percellome database.
1.Aisaki KI, Ono R, Kanno J, Kitajima S. [Percellome Project: research on molecular mechanisms of toxicological responses based on transcriptomics and epigenetics]. Nihon Yakurigaku Zasshi. 2022;157(3):200-206. Japanese. https://doi.org/10.1254/fpj.21122
2.Takeshi Hase, Samik Ghosh, Ken-ichi Aisaki, Satoshi Kitajima, Jun Kanno, Hiroaki Kitano, Ayako Yachie, DTox: A deep neural network-based in visio lens for large scale toxicogenomics data, The Journal of Toxicological Sciences, 2024;9(3), 105-115 https://doi.org/10.2131/jts.49.105