Large-scale comparison of machine learning methods for drug target prediction on ChEMBL

Mayr, Andreas; Klambauer, Günter; Unterthiner, Thomas; Steijaert, Marvin; Wegner, Jörg K.; Ceulemans, Hugo; Clevert, Djork-Arné; Hochreiter, Sepp

doi:10.1039/c8sc00148k

Cited by 443 publications

(536 citation statements)

References 52 publications

(53 reference statements)

Supporting

Mentioning

493

Contrasting

Order By: Relevance

“…[206] Current achievements in reaction prediction and retrosynthesis demonstrate the potential of machine learning to solve one of the bottlenecks of state-of-the-art materials design, which is the planning of efficient reaction routes of new molecules. Examples can be given for, e.g., feature detection, [207] bioactivity prediction, [208] or drug target prediction, [209] and others. Examples can be given for, e.g., feature detection, [207] bioactivity prediction, [208] or drug target prediction, [209] and others.…”

Section: Discussionmentioning

confidence: 99%

Toward Design of Novel Materials for Organic Electronics

et al. 2019

View full text Add to dashboard Cite

organic/inorganic perovskites solar cells [6] to printable electronic circuits based on organic field-effect transistors (OFETs). [7] While OLED displays outperform their inorganic counterparts in terms of energy efficiency, [8] scientific and technical challenges concerning the stability and processability of the organic materials used in large-area OLEDs and organic solar cells remain. New challenges arise from applications, such as displays on flexible substrates, OLED lightning, large area displays as well as for printable or solution processable larger area solar cells. [8] Many of the remaining challenges are material related, e.g., the low mobility of charge carriers in organic materials in general and in amorphous organic semiconductors in particular. There are other materials related issues, such as limited OLED life times due to unstable blue hosts and emitters, [9] low fill factors, and therefore reduced power conversion efficiencies of organic solar cells, [10,11] low conductivity and high costs of organic charge transport layers of perovskite solar cells [6,12,13] and low conductivity and hard processability of crystalline OFET materials. [14] Conductivity and injection can be improved by doping the organic thin films with molecular dopants with high electron affinities (p-type) [15] or low ionization energies (n-type). [16] The doping mechanism of organic materials is in many cases not well understood, [17] making material and device optimization a costly experimental endeavor.The development of better materials is presently based on chemical insight, in part guided by theoretical understanding, or the experimental screening of large numbers of compounds. Given the size of the potentially available chemical space this remains a costly and time-consuming approach. Recent successes in experimental design of novel materials and concepts include the development of a stable strong molecular n-type dopant, [16] a study about the quantitative relation between interaction parameter, miscibility, and function of conjugated polymer donors and small-molecule acceptors for bulk heterojunctions as used in organic solar cells [18] the development of a universal strategy for ohmic hole injection into organic semiconductors with high ionization energies [19] and many others. [20][21][22][23][24][25] Another example is the development of strategies to harvest triplet excitons in organic light emitting diodes. For this purpose, novel classes of emitter molecules were developed, which include thermally activated delayed fluorescence (TADF)-based molecules, [26][27][28][29][30][31][32] rotationally accessed spin-state inversion, [33] and radical-based emitters. [34] Materials for organic electronics are presently used in prominent applications, such as displays in mobile devices, while being intensely researched for other purposes, such as organic photovoltaics, large-area devices, and thin-film transistors. Many of the challenges to improve and optimize these applications are material related and there is a nearly inf...

show abstract

Section: Discussionmentioning

confidence: 99%

Toward Design of Novel Materials for Organic Electronics

et al. 2019

View full text Add to dashboard Cite

show abstract

“…They are typically used to produce libraries of potential drugs. This field has greatly benefited from recent advances in deep learning technology [11,15,39], combined with the large amounts of ligand data available [18,30]. However, with ligand-based methods, ligand-to-target interactions, which ultimately determine activity, are rarely modelled.…”

Section: Computational Drug Discoverymentioning

confidence: 99%

Augmented base pairing networks encode RNA-small molecule binding preferences

Oliver

Mallet

Gendron

et al. 2019

Preprint

View full text Add to dashboard Cite

Recent studies have identified small RNA-binding molecules with potential for therapeutic applications including novel antibiotics, antivirals, protein synthesis controllers, and CRISPR activators. As RNA binding data accumulates, machine learning methods are becoming attractive approaches to accelerate the discovery of RNA-targeting drugs. While atomic level representations of molecular systems are regarded as a critical step for physics-based and data-driven drug design, the singularity and hierarchical organization of RNA structures challenges this paradigm. In this work, we present a machine learning framework for assisting in the discovery of small RNA-binding molecules from graphical representations of RNA structures. A key feature of our tool is that it does not rely on manual feature engineering or costly physical simulations. Instead, we extract molecule-binding information directly from known RNA-ligand complexes by combining graph embedding methods with machine learning models. Given only the graph representation of an RNA site as input, our tool is able to reliably predict chemical features, also known as fingerprints, of the observed ligands. The resulting fingerprints can be used to classify RNA sites according to their binding preferences, screen databases for promising ligands, or act as constraints for generating novel ligands. We show consistent performance for across ligand classes in enriching the known binder from a larger ligand library. As a validation, we applied this method to successfully identify binding residues in unbound RNA structures for three different riboswitches. These results also suggest that small molecule binding preferences in RNA can be extracted directly from the nucleotide pairing level.

show abstract

“…The attention weights were obtained after training the models using the binding affinity labels, that is, without extra supervision from the pairwise interaction labels. The clustering-based cross-validation procedure [33] was used during the training process, which ensures that similar compounds (or/and proteins) in the same clusters were not shared between training and test sets. Three cross-validation settings were used in the evaluation, including the new-compound setting, in which the test compounds were never seen in the training process, the new-protein setting, in which the test proteins were never seen in the training data, and the both-new cross-validation setting, in which both compounds and proteins in the test data were never seen during training.…”

Section: Systematic Evaluation Of the Interpretability Of Neural Attementioning

confidence: 99%

MONN: a Multi-Objective Neural Network for Predicting Pairwise Non-Covalent Interactions and Binding Affinities between Compounds and Proteins

Wan

Shu

et al. 2019

Preprint

View full text Add to dashboard Cite

Computational approaches for inferring the mechanisms of compound-protein interactions (CPIs) can greatly facilitate drug development. Recently, although a number of deep learning based methods have been proposed to predict binding affinities and attempt to capture local interaction sites in compounds and proteins through neural attentions, they still lack a systematic evaluation on the interpretability of the identified local features. In addition, in these previous approaches, the exact matchings between interaction sites from compounds and proteins, which are generally important for understanding drug mechanisms of action, still remain unknown. Here, we compiled the first benchmark dataset containing the inter-molecular non-covalent interactions for more than 10,000 compound-protein pairs, and used it to systematically evaluate the interpretability of neural attentions in existing prediction models. We developed a multi-objective neural network, called MONN, to predict both non-covalent interactions and binding affinity for a given compound-protein pair. MONN uses convolution neural networks on molecular graphs of compounds and primary sequences of proteins to effectively capture the intrinsic features from both inputs, and also takes advantage of the predicted non-covalent interactions to further boost the accuracy of binding affinity prediction. Comprehensive evaluation demonstrated that while the previous neural attention based approaches fail to exhibit satisfactory interpretability results without extra supervision, MONN can successfully predict non-covalent interactions on our benchmark dataset as well as another independent dataset derived from the Protein Data Bank (PDB). Moreover, MONN can outperform other state-of-the-art methods in predicting compound-protein binding affinities. In addition, the pairwise interactions predicted by MONN displayed compatible and accordant patterns in chemical properties, which provided another evidence to support the strong predictive power of MONN. These results suggested that MONN can offer a powerful tool in predicting binding affinities of compound-protein pairs and also provide useful insights into understanding the molecular mechanisms of compound-protein interactions, which thus can greatly advance the drug discovery process. The source code of the MONN model and the dataset creation process can be downloaded from https://github.com/lishuya17/MONN.

show abstract

Large-scale comparison of machine learning methods for drug target prediction on ChEMBL

Abstract: The to date largest comparative study of nine state-of-the-art drug target prediction methods finds that deep learning outperforms all other competitors. The results are based on a benchmark of 1300 assays and half a million compounds.

Cited by 443 publications

References 52 publications

Toward Design of Novel Materials for Organic Electronics

Toward Design of Novel Materials for Organic Electronics

Augmented base pairing networks encode RNA-small molecule binding preferences

MONN: a Multi-Objective Neural Network for Predicting Pairwise Non-Covalent Interactions and Binding Affinities between Compounds and Proteins

Contact Info

Product

Resources

About