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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Deep neural networks have achieved state of the art accuracy at classifyingmolecules with respect to whether they bind to specific protein targets. A keybreakthrough would occur if these models could reveal the fragmentpharmacophores that are causally involved in binding. Extracting chemicaldetails of binding from the networks could potentially lead to scientificdiscoveries about the mechanisms of drug actions. But doing so requires shininglight into the black box that is the trained neural network model, a task thathas proved difficult across many domains. Here we show how the bindingmechanism learned by deep neural network models can be interrogated, using arecently described attribution method. We first work with carefully constructedsynthetic datasets, in which the 'fragment logic' of binding is fully known. Wefind that networks that achieve perfect accuracy on held out test datasetsstill learn spurious correlations due to biases in the datasets, and we areable to exploit this non-robustness to construct adversarial examples that foolthe model. The dataset bias makes these models unreliable for accuratelyrevealing information about the mechanisms of protein-ligand binding. In lightof our findings, we prescribe a test that checks for dataset bias given ahypothesis. If the test fails, it indicates that either the model must besimplified or regularized and/or that the training dataset requiresaugmentation.

Using Attribution to Decode Dataset Bias in Neural Network Models for Chemistry