DeepTox: Toxicity Prediction using Deep Learning

By Andreas Mayr, Günter Klambauer, Thomas Unterthiner, Sepp Hochreiter. 2016

921 citations recorded by the source corpus.

The Tox21 Data Challenge has been the largest effort of the scientific community to compare computational methods for toxicity prediction. This challenge comprised 12,000 environmental chemicals and drugs which were measured for 12 different toxic effects by specifically designed assays. We participated in this challenge to assess the performance of Deep Learning in computational toxicity prediction. Deep Learning has already revolutionized image processing, speech recognition, and language understanding but has not yet been applied to computational toxicity. Deep Learning is founded on novel algorithms and architectures for artificial neural networks together with the recent availability of very fast computers and massive datasets. It discovers multiple levels of distributed representations of the input, with higher levels representing more abstract concepts. We hypothesized that the construction of a hierarchy of chemical features gives Deep Learning the edge over other toxicity prediction methods. Furthermore, Deep Learning naturally enables multi-task learning, that is, learning of all toxic effects in one neural network and thereby learning of highly informative chemical features. In order to utilize Deep Learning for toxicity prediction, we have developed the DeepTox pipeline. First, DeepTox normalizes the chemical representations of the compounds. Then it computes a large number of chemical descriptors that are used as input to machine learning methods. In its next step, DeepTox trains models, evaluates them, and combines the best of them to ensembles. Finally, DeepTox predicts the toxicity of new compounds. In the Tox21 Data Challenge, DeepTox had the highest performance of all computational methods winning the grand challenge, the nuclear receptor panel, the stress response panel, and six single assays (teams ``Bioinf@JKU''). We found that Deep Learning excelled in toxicity prediction and outperformed many other computational approaches like naive Bayes, support vector machines, and random forests.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

In Materials Chemistry

In Molecular Biology

In Safety, Robustness & Interpretability

Citations in the corpus

Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs) (2015, background): “ and allowed to detect biclusters corresponding to biological modules in gene expression datasets (Clevert et al., 2015) and to identify toxicophores in toxicity prediction (Unterthiner et al., 2015; Mayr et al., 2015). The enabling features for these interpretations is that activation can be clearl”
Molecular graph convolutions: moving beyond fingerprints (2016, methodology): “In particular, multitask neural networks have produced consistent gains relative to baseline models such as random forest and logistic regression [6, 17, 19, 29].”
Opportunities and obstacles for deep learning in biology and medicine (2017, methodology): “Many studies employed multi-task learning to predict chemical bioactivity [299,302] and drug toxicity [303,421].”
Using Rule-Based Labels for Weak Supervised Learning: A ChemNet for Transferable Chemical Property Prediction (2017, methodology): “Recent work have also started using deep neural networks (DNN) [9, 19, 24, 27], that are on average, typically more accurate than traditional ML models [12, 14].”
Self-Normalizing Neural Networks (2017, background): “In 2015, the challenge organized by the US NIH was won by an ensemble of shallow ReLU FNNs which achieved an AUC of 0.846 [23].”
Molecular de-novo design through deep reinforcement learning (2017, neutral): “In order to decrease the nearest neighbour similarity between the training and testing structures [34–36], the actives were grouped in clusters based on their molecular similarity.”
Computational Protein Design with Deep Learning Neural Networks (2018, methodology): “…Particularly in computational biology/chemistry, it has been used in protein-ligand scoring, 40-42 protein-protein interaction prediction, 43 protein secondary structure prediction, 44-49 protein contact map prediction, 50-52 and compound toxicity 53,54 and liver injury prediction, 55 among others.”
Chemception: A Deep Neural Network with Minimal Chemistry Knowledge Matches the Performance of Expert-developed QSAR/QSPR Models (2017, methodology): “These human-expert features are then used by machine learning algorithms for QSAR and QSPR modeling, which in recent years have also included deep neural networks.(9-23) However, we observed that even in the most recent chemistry literature, deep learning in chemistry still falls under this traditio”
A guide to machine learning for biologists (2021, background): “Drug toxicity prediction [54] Can fit datasets with fewer layers than architectures such as CNNs, making them easier and faster to train Easy to overfit”
ChemNet: A Transferable and Generalizable Deep Neural Network for Small-Molecule Property Prediction (2017, background): “Recent advances in the field have also started using deep neural networks (DNN) on the same engineered features to predict a wide range of small-molecule properties [2, 3, 4, 5].”
ProTox-II: a webserver for the prediction of toxicity of chemicals (2018, methodology): “Several computational models have been developed to predict the toxicity of chemicals (8,10,11).”
SMILES2Vec: An Interpretable General-Purpose Deep Neural Network for Predicting Chemical Properties (2017, background): “More recently, deep learning (DL) models have also been developed [9, 21, 28, 31].”
Augmenting large language models with chemistry tools (2023, background): “…expert-designed artificial intelligence (AI) systems that tackle specific problems, such as reaction prediction 16–20 , retrosynthesis planning 21–27 , molecular property prediction 28–32 , de-novo molecular generation 33,34 , materials design 35,36 and, more recently, Bayesian Optimization 37–39 .”