Case Study: Leading Chemical Service Provider achieves 84.1% AUC toxicity prediction with Quantiphi's deep learning virtual testing solution

Toxicity/Quality prediction using Deep learning

Leading Chemical Service Provider partnered with Quantiphi to improve virtual testing for chemical toxicity and quality prediction. The customer needed a more reliable way to assess the effects of chemical compounds on human health, since current experiments are time- and cost-intensive and existing computational approaches were not accurate enough.

Quantiphi implemented a deep learning-based virtual testing pipeline using Python and TensorFlow, built around a stack of self-normalizing networks with a multitask cost function. The solution achieved a state-of-the-art AUC score of 84.1% across the Tox21 toxicity markers and helped filter out risky compounds, reducing dependence on in-vitro experiments.