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

By Andreas Mayr, Günter Klambauer, Thomas Unterthiner, M. Steijaert, Jörg Kurt Wegner, Hugo Ceulemans, Djork-Arné Clevert, Sepp Hochreiter. 2018

442 citations recorded by the source corpus.

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.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology, Computational Chemistry & Drug Discovery

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

1. Francesca Grisoni
2. Gisbert Schneider
3. Tommi S. Jaakkola
4. Marinka Zitnik
5. Regina Barzilay
6. Jure Leskovec
7. Kyle Swanson
8. Kenneth Atz

In Materials Chemistry

1. Gisbert Schneider
2. Francesca Grisoni
3. Jure Leskovec
4. Kyle Swanson
5. Tommi S. Jaakkola
6. Regina Barzilay
7. Marinka Zitnik
8. Sepp Hochreiter

In Molecular Biology

1. Tommi S. Jaakkola
2. Marinka Zitnik
3. Regina Barzilay
4. Kyle Swanson
5. Jure Leskovec
6. Vijay S. Pande
7. Sepp Hochreiter
8. Weihua Hu

In Computational Chemistry & Drug Discovery

1. Kyle Swanson
2. Tommi S. Jaakkola
3. Regina Barzilay
4. Sepp Hochreiter
5. Gisbert Schneider
6. Günter Klambauer
7. Jure Leskovec
8. Thomas Unterthiner

Citations in the corpus

• Strategies for Pre-training Graph Neural Networks (2019, methodology): “For graph-level multi-task supervised pre-training, we use a preprocessed ChEMBL dataset (Mayr et al., 2018; Gaulton et al., 2011 Dataset splitting.”
• Modern Hopfield Networks for Few- and Zero-Shot Reaction Prediction (2021, methodology): “One could for example use raw fingerprints, graph neural networks (Gilmer et al., 2017) or SMILES/SMARTS-based RNNs (Mayr et al., 2018) or Transformers.”
• Message passing neural networks for molecular property prediction (2019, methodology): “Improvements in property prediction accuracy using these models has primary come through the development of better molecular descriptors [44, 8, 15, 33, 31, 32].”
• Few-shot Conformal Prediction with Auxiliary Tasks (2021, methodology): “We use the ChEMBL dataset (Mayr et al., 2018), and regress the pChEMBL value (a normalized log-activity metric) for individual molecule-property pairs.”
• Fréchet ChemblNet Distance: A metric for generative models for molecules (2018, methodology): “For this purpose we used a ChEMBL molecule clustering which was produced in a previous work [13].”
• MolTrans: Molecular Interaction Transformer for drug–target interaction prediction (2020, background): “ssication problem, and make prediction by feeding the inputs through deep learning models such as deep neural network (DNN) [3], deep belief network (DBN) [4], and convolutional neural network (CNN) [5, 6, 7]. Despite these eorts, the following challenges are still open. 1. Inadequate modeling of in”
• A Deep Learning Approach to Antibiotic Discovery. (2020, unknown): “Decades of prior work in chemoinformatics have developed models for molecular property prediction (Mayr et al., 2018; Wu et al., 2017).”
• Geometric deep learning on molecular representations (2021, background): “Transformers String notation (encoded as a graph) Prediction of properties in drug discovery [51], synthesis planning [46, 52], prediction of reaction yields [53], generative molecular design [54] and protein structure prediction [6, 7]”
• Pre-training Graph Neural Networks (2019, methodology): “…molecular graphs, we use 2 million unlabeled molecules sampled from the ZINC15 database [47] for the self-supervised pre-training, and use a preprocessed ChEMBL dataset [29, 12], containing 456K molecules with 1310 kinds of binary biochemical assays, for the graph-level supervised pre-training.”