The model uses a combination of various variants of neural networks, the total prediction accuracy of which was 93%, while the individual classifiers included in the model reached 87-88%. It is also the upper precision limit for currently available data, since the model uncertainty is dominated by aleatoric input, i.e. irreducible, inherent to the data used. The results were published in the August issue of Nature Machine Intelligence under the title: “Minimal-uncertainty prediction of general drug-likeness based on Bayesian neural networks”.
More information: https://www.nature.com/articles/s42256-020-0209-y
Congratulations and we wish you all the success in your research!