Axiom Bio has released a comprehensive dataset of liver toxicity profiles for AI-based drug safety assessments. The dataset includes high‑content imaging data from human hepatocyte cultures exposed to around 130,000 compounds. Each compound is accompanied by multi-channel fluorescence microscopy images and annotated phenotypic readouts reflecting liver cell health oai_citation:0‡techlifesci.com.
Key Developer Features
- Multi‑label classification data for hepatotoxic vs non‑toxic responses
- High‑resolution cellular images with quantifiable phenotypic features
- Standardized metadata for compound concentration, exposure time, and assay conditions
- Exportable formats including CSV and NumPy arrays for seamless integration
Example Usage
from axiombio import HepatoDataset
data = HepatoDataset("axiom_livertox_130k")
img, label = data[123]
print(label, img.shape)
The toolkit also includes helper methods for data normalization, visualization, stratified train/test splitting, and integration with deep learning frameworks like PyTorch and TensorFlow.
Why It Matters
Drug‑induced liver injury is a leading cause of clinical trial failure and post‑market drug withdrawal. By providing a large, annotated dataset, Axiom Bio enables AI models to learn from biologically realistic images, potentially improving early safety prediction and reducing late‑stage drug failure.
Developers can fine‑tune convolutional neural networks or train explainable models to classify toxicity outcomes, perform transfer learning, or extract toxicity‑related features for downstream analyses.
What’s Next
Axiom Bio is organizing a community challenge for AI teams to benchmark their toxicity prediction models. The dataset is currently available via the company’s portal, with academic access granted free of charge.
Sources
https://www.techlifesci.com/p/weekly-techbio-highlights-45-biotech
Top comments (0)