machine learning algorithm predicts toxicity better than animal testing

60 seconds science podcast episode Computerized Chemical Toxicity Prediction Beats Animal Testing
discusses an algorithmic way to find reaction to chemical substances instead of running animal test. The idea is to consider all test that has been done historically and based on chemical similarity between compounds the algorithm predicts the reaction.


The article claims to be un-biased but these algorithms are biased by such simple issues as similarity measure or the way the model is validated. In addition, the actual paper is less focused on individual prediction but is more about the sexy area of network topology of the space of chemicals.


written on 2018/08/13