Industrial ADME data Sets and Deep Learning #Chemoinformatcs

In pharma, predictive models are widely used such as activity, ADME, phychem and Tox. And recently many articles which use Deep Learning are published. Deep Learning is powerful tool for prediction but it is difficult to find appropriate hyper parameters. Today I read informative article published researcher from Lilly.
https://pubs.acs.org/doi/10.1021/acs.jcim.8b00671

They investigated performance of DNN and SVM with in-house ADMET datasets such as solubility, CYP inhibition, p-gp efflux, MDCK, Metabolic stability and fu.

The author used ECFP6 as both input and DNN is implemented by using pytorch (code and dataset is not disclosed). And investigated hyper parameters are learning rate, weight decay, dropout rate, activation function and batch normalization. And they evaluated their performance with AUROC, MCC for classification and RMSE for regression.

It is worth for me that batch normalization shows big impact for reducing the potential of poor performance. It could find in Fig1.

Next they analyzed the effect of the hyper parameters on regression models. In regression models, they found that interaction of activation function and learning rate is most important. And many other results are described in the article. If reader who is interested the article I recommend to read it.

Finally it is interesting for me that DNN which has optimized hyper parameters showed marginal improvement over the SVM models.

For prediction DNN does not big advantage compared to traditional machine learning approach such as RF, SVM etc. But it has attractive feature like auto encoding, transfer learning, generator etc etc.

I need study and practice not only DNN but also traditional ML approach.

Published by iwatobipen

I'm medicinal chemist in mid size of pharmaceutical company. I love chemoinfo, cording, organic synthesis, my family.

One thought on “Industrial ADME data Sets and Deep Learning #Chemoinformatcs

  1. I read your blog. This time of composed drug disclosure helps drug revelation CROs to choose and assess the suitability of the different prescription contenders. ADME organizations help in concluding the basic pieces of the drug compounds on the going with limits. The provided information is very useful for ADME services. Keep continuing to post further.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: