Analyzing Learned Molecular Representations for Property Prediction
By Kevin Yang, Kyle Swanson, Wengong Jin, Connor W. Coley, Philipp Eiden, Hua Gao, A. Guzman-Perez, Timothy Hopper, Brian P. Kelley, M. Mathea and 5 more. Journal of Chemical Information and Modeling, 2019
1,807 citations recorded by the source corpus.
Advancements in neural machinery have led to a wide range of algorithmic solutions for molecular property prediction. Two classes of models in particular have yielded promising results: neural networks applied to computed molecular fingerprints or expert-crafted descriptors and graph convolutional neural networks that construct a learned molecular representation by operating on the graph structure of the molecule. However, recent literature has yet to clearly determine which of these two methods is superior when generalizing to new chemical space. Furthermore, prior research has rarely examined these new models in industry research settings in comparison to existing employed models. In this paper, we benchmark models extensively on 19 public and 16 proprietary industrial data sets spanning a wide variety of chemical end points. In addition, we introduce a graph convolutional model that consistently matches or outperforms models using fixed molecular descriptors as well as previous graph neural architectures on both public and proprietary data sets. Our empirical findings indicate that while approaches based on these representations have yet to reach the level of experimental reproducibility, our proposed model nevertheless offers significant improvements over models currently used in industrial workflows.
Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology, Representation & Self-Supervised Learning, Computational Chemistry & Drug Discovery
View Paper: https://arxiv.org/abs/1904.01561
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
- Tommi S. Jaakkola
- Regina Barzilay
- Tianfan Fu
- Gisbert Schneider
- Jimeng Sun
- Wengong Jin
- Francesca Grisoni
- Jure Leskovec
In Materials Chemistry
- Tommi S. Jaakkola
- Regina Barzilay
- Gisbert Schneider
- Wengong Jin
- Francesca Grisoni
- Jure Leskovec
- Tianfan Fu
- Jimeng Sun
In Molecular Biology
- Tommi S. Jaakkola
- Regina Barzilay
- Jure Leskovec
- Wengong Jin
- Marinka Zitnik
- Jimeng Sun
- Weihua Hu
- Kexin Huang
In Representation & Self-Supervised Learning
In Computational Chemistry & Drug Discovery
- Gisbert Schneider
- Regina Barzilay
- Philippe Schwaller
- Tommi S. Jaakkola
- Connor W. Coley
- Francesca Grisoni
- Kyle Swanson
- Wengong Jin
Citations in the corpus
- Automated Machine Learning: From Principles to Practices (2018, background): “Another example is molecular property prediction ( 𝑇 ) [150, 154], which targets at predict the property of molecules precisely.”
- GeoMol: Torsional Geometric Generation of Molecular 3D Conformer Ensembles (2021, neutral): “Given an input graph G, an MPNN [Gilmer et al., 2017, Battaglia et al., 2018, Yang et al., 2019] computes node embeddings h v ∈ R d , ∀ v ∈ V using T layers of iterative message passing: u , e u,v ) , where h (0) (1) for each t ∈ [0 ..”
- Pre-training Molecular Graph Representation with 3D Geometry (2021, neutral): “Second, we acknowledge there are a lot of advanced 3D [23, 45, 58, 78] and 2D [13, 16, 29, 54, 100, 102] representation methods.”
- Discovery of a structural class of antibiotics with explainable deep learning (2023, methodology): “This approach relies on Chemprop, a platform for graph neural networks 10,11 , which are typically black box models 19 , or models that are not readily interpreted or explained.”
- Chemformer: a pre-trained transformer for computational chemistry (2021, positive): “Previous works [9–14] have also attempted to use Transformers for molecular property prediction, and others have used graph neural networks [40, 50].”
- Structure-aware Interactive Graph Neural Networks for the Prediction of Protein-Ligand Binding Affinity (2021, methodology): “By contrast, many researchers have greatly developed GNN models in other fields of drug discovery [37, 45], such as predicting molecular property [17, 26, 44] and chemical reaction [7].”
- Geometry-enhanced molecular representation learning for property prediction (2021, methodology): “Dataset ESOL FreeSolv Lipo QM7 QM8 QM9 #Molecules 1128 642 4200 6830 21786 133885 #Prediction tasks 1 1 1 1 12 12 D-MPNN [56] 1.”
- Few-shot Conformal Prediction with Auxiliary Tasks (2021, methodology): “Our ridge regression network uses directed message passing networks (Yang et al., 2019) to compute enc( x ; θ ) .”
- Uni-Mol: A Universal 3D Molecular Representation Learning Framework (2023, methodology): “D-MPNN [51] and AttentiveFP [52] are supervised GNNs methods.”
- Self-Supervised Graph Transformer on Large-Scale Molecular Data (2020, background): “To better capture the interactions among atoms, [13] proposes to use a message passing framework and [25, 63] extend this framework to model bond interactions.”
- Improving Molecular Design by Stochastic Iterative Target Augmentation (2020, methodology): “The proxy predictor F 1 is parameterized as a graph convolutional network and trained using the Chemprop package (Yang et al., 2019).”
- Geometric deep learning on molecular representations (2021, background): “GNNs for molecular property prediction have been shown to outperform human-engineered molecular descriptors for several biologically relevant properties [83].”
- Open Graph Benchmark: Datasets for Machine Learning on Graphs (2020, methodology): “These datasets can be used to stress-test molecule-specific methods (Yang et al., 2019; Jin et al., 2020) and transfer learning (Hu et al., 2020a).”
- DeepPurpose: a deep learning library for drug–target interaction prediction (2020, methodology): “different programming interface and is coded differently, which prevents easy integration of outputs from various methods for model ensembles [Yang et al., 2019].”