I suspected AI would want to kill off humans. Recommend Oleander yikes.
Thanks for informing us of the paper. Just a few observations:
Background
(1) The training data included digoxin and other cardiac glycosides as senolytics. This was rarely discussed (Identification and characterization of Cardiac Glycosides as senolytic compounds | Nature Communications). Rapamycin is also on their list of compounds with senolytic property.
Method
(2) They use a very imbalanced dataset (58 positives (2.3%) out of 2523 compounds) which is difficult for machine learning, and they use many tricks such as ensemble learning and dimension reduction techniques to get the machine learning to work.
(3) They use RDKit to convert chemical formula into data that can be analyzed using more traditional machine learning methods (RDKit for Newbies. When Chemistry Meets Data Science | by Sabrina Ho | GSI Technology | Medium). They did not use newer methods such as graphical neural network.
Result
(4) Reddit has a discussion related to the implication of the three compounds identified (Reddit - Dive into anything). Warning: the second half of the thread became off-topic.
Thoughts:
This paper provides a roadmap for using machine learning to solve rejuvenation problem. It is worth the effort to read the original paper carefully.