Publication
PNAS
Stage
Journal article
What we read
Summary of the abstract
Authors
Haoyue Tan, Tong Bao, Yin Fang, Rong Zhang, Jinsha Jin, Dan Xu, Lan Xie, Huan Zhong, Xuezhi Xiao, Huixiao Hong, Emilio Benfenati, Tadahaya Mizuno, Qing Zhou, Jingfan Qiu, Changsheng Qu, Yan Mao, Xiangyi Yu, Jing Guo, Hongxia Yu, Xiaowei Zhang, Wei Shi
Universities and research institutions
Nanjing University, Zhejiang University, Ministry of Ecology and Environment, National Center for Toxicological Research, United States Food and Drug Administration, Mario Negri Institute for Pharmacological Research, The Institute of Statistical Mathematics, Statistical Service

Institution metadata: OpenAlex record ↗

What they did and found

Problem: predicting many toxic effects is hard. Approach: use a cause-and-effect map from chemical shape to possible outcomes, then test how well it flags risks. Finding: the method rests on a simple model with limited real-world checks.

Why it matters

For safety teams, looking at several risks at once can help screening, but results aren’t proven in real use and should be treated with care.

The idea links how a chemical looks to how it might affect several biological targets, all tied together by one cause-and-effect map.

In practice, an example would compare two new substances to see if they share disruption patterns, guiding checks while real tests could give different results.

What remains uncertain

Abstract-only work with no real-world proof yet; findings may change as new data arrives.

Original sources · 1
  1. Causality-integrated graph learning for multi-endpoint toxicity prediction ↗PNAS · 2026-10-05

Check the original paper for its authors, methods, version and access terms.