Original title: FuTCM-PDD: a multi-level information fusion framework integrating KAN modeling and AutoML for phenotype-based drug discovery from Traditional Chinese Medicine
Zewen Wang, Yanxia Liu, Yue Ren, Qun Li +7 · September 2026
Finding new medicines by looking at their real-world bodily effects, called phenotypes, is a powerful way to discover treatments. Traditional Chinese medicine focuses heavily on these bodily effects and herb traits, but connecting herbs to specific biological targets is difficult. To address this, researchers built an AI framework called FuTCM-PDD. It works by connecting herbs, their chemical compounds, biological targets, pathways, bodily effects, and traditional properties into one massive map. The team used a neural network design called a Kolmogorov-Arnold Network to predict how compounds bind to targets. They also tested 168 machine learning algorithms to connect pharmacological effects to traditional herb properties and efficacies. The system revealed that an herb called Qianghuo, or Notopterygii Rhizoma et Radix, along with its active compounds, has a lipid-lowering effect. Laboratory tests on a liver cell model of fat buildup confirmed this finding, showing that the system finds candidates conventional methods often overlook.
Imagine you have a giant cookbook with thousands of recipes, but each recipe only lists a few ingredients and a vague note like 'good for cold days.' You want to know which recipe will cure your friend's cough, but the book never says so directly. You could guess from the notes, but you'd keep circling the same few famous recipes. In this paper, a computer reads all the recipes, the ingredients, and the notes together, then finds a connection no one noticed: a herb usually used for one thing also treats fatty liver.
Drug discovery often keeps circling the same well-known compounds. This framework instead links a herb's traditional properties to its molecular targets and effects, and it experimentally confirmed a new lipid-lowering use for an old herb. That means a faster, cheaper route to first-in-class drugs and a way past recycled associations.
The abstract reports only one experimental validation, an in vitro liver-cell model, so the findings have not been shown to work in living animals or humans. The DTI improvement is measured only by Recall@10, not by other metrics. The framework's predictions remain largely computational, and the abstract does not report how many other predictions were tested.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
FuTCM-PDD: a multi-level information fusion framework integrating KAN modeling and AutoML for phenotype-based drug discovery from Traditional Chinese Medicine — Zewen Wang, Yanxia Liu, Yue Ren et al. — Chinese Medicine — doi:10.1186/s13020-026-01490-1
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.