English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Artificial intelligence, trained via machine learning or computationalstatistics algorithms, holds much promise for the improvement of small moleculedrug discovery. However, structure-activity data are high dimensional with lowsignal-to-noise ratios and proper validation of predictive methods isdifficult. It is poorly understood which, if any, of the currently availablemachine learning algorithms will best predict new candidate drugs. 25 publiclyavailable molecular datasets were extracted from ChEMBL. Neural nets, randomforests, support vector machines (regression) and ridge regression were thenfitted to the structure-activity data. A new validation method, based onquantile splits on the activity distribution function, is proposed for theconstruction of training and testing sets. Model validation based on randompartitioning of available data favours models which overfit and `memorize' thetraining set, namely random forests and deep neural nets. Partitioning based onquantiles of the activity distribution correctly penalizes models which canextrapolate onto structurally different molecules outside of the training data.This approach favours more constrained models, namely ridge regression andsupport vector regression. In addition, our new rank-based loss functions giveconsiderably different results from mean squared error highlighting thenecessity to define model optimality with respect to the decision task at hand.Model performance should be evaluated from a decision theoretic perspectivewith subjective loss functions. Data-splitting based on the separation of highand low activity data provides a robust methodology for determining the bestextrapolating model. Simpler, traditional statistical methods such as ridgeregression outperform state-of-the-art machine learning methods in thissetting.

A decision theoretic approach to model evaluation in computational drug discovery