Related Experiment Video
Updated: Jul 1, 2025
12:41
Multiparametric Tumor Organoid Drug Screening Using Widefield Live-Cell Imaging for Bulk and Single-Organoid Analysis
Published on: December 23, 2022
4.8K
DTox: A deep neural network-based in visio lens for large scale toxicogenomics data
Takeshi Hase1,2,3,4,5, Samik Ghosh1, Ken-Ichi Aisaki6
1The Systems Biology Institute, Saisei Ikedayama Bldg.
The Journal of Toxicological Sciences
|March 3, 2024
Summary
DTox, a novel deep neural network approach, analyzes toxicogenomic data by learning from image representations of gene expression. This in visio method enhances the identification of significant genes, outperforming traditional statistical and machine learning techniques.
Area of Science:
- Toxicogenomics
- Bioinformatics
- Computational Biology
Background:
- Large-scale omics data, especially transcriptomics, are crucial for safety pharmacology and chemical risk assessment.
- Traditional bioinformatics methods struggle with the high dimensionality of toxicogenomic data (time, dose, gene expression).
- Expert visual inspection aids in identifying significant genes but is time-consuming.
Purpose of the Study:
- To develop an efficient and accurate method for analyzing multidimensional toxicogenomic data.
- To leverage deep learning for interpreting complex gene expression patterns.
- To create an "in visio" approach that mimics expert visual analysis.
Main Methods:
- Developed DTox, a deep neural network-based "in visio" approach.
- Trained DTox on image representations (3D surface plots) of gene expression data from the Percellome database.
- Used expert-labeled gene probe significance for training, rather than raw numerical values.
- Integrated explainability modules to interpret model decision-making.
Main Results:
- DTox successfully learned from image representations of dose-time gene expression data.
- The DTox model outperformed traditional statistical and numerical machine learning methods in identifying significant genes.
- Explainability modules revealed insights into the visual analysis process employed by human experts.
- Demonstrated the efficacy of image-driven neural networks over numerical approaches.
Conclusions:
- DTox offers a powerful alternative for analyzing complex toxicogenomic data, overcoming limitations of traditional methods.
- The "in visio" deep learning approach enhances efficiency and accuracy in safety pharmacology and risk assessment.
- The model's explainability provides a window into expert visual analysis, aiding scientific understanding.
- DTox has potential for broader applications in analyzing multidimensional numerical data across medical data sciences.
