ATOM3D: Tasks On Molecules in Three Dimensions

By Raphael J. L. Townshend, M. Vögele, Patricia Suriana, Alexander Derry, A. Powers, Yianni Laloudakis, S. Balachandar, Brandon M. Anderson, Stephan Eismann, Risi Kondor and 2 more. 2020

173 citations recorded by the source corpus.

Computational methods that operate directly on three-dimensional molecular structure hold large potential to solve important questions in biology and chemistry. In particular deep neural networks have recently gained significant attention. In this work we present ATOM3D, a collection of both novel and existing datasets spanning several key classes of biomolecules, to systematically assess such learning methods. We develop three-dimensional molecular learning networks for each of these tasks, finding that they consistently improve performance relative to one- and two-dimensional methods. The specific choice of architecture proves to be critical for performance, with three-dimensional convolutional networks excelling at tasks involving complex geometries, while graph networks perform well on systems requiring detailed positional information. Furthermore, equivariant networks show significant promise. Our results indicate many molecular problems stand to gain from three-dimensional molecular learning. All code and datasets can be accessed via this https URL .

How researchers see this paper

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

Citations in the corpus

TANKBind: Trigonometry-Aware Neural NetworKs for Drug-Protein Binding Structure Prediction (2022, background): “Similar to previous methods [23, 44], we predict negative log-transformed binding affinity.”
Learning inverse folding from millions of predicted structures (2022, methodology): “We evaluate models on zero-shot prediction of mutational effects on protein complex interfaces, using the Atom3D benchmark (Townshend et al., 2020) which incorporates binding free energy changes in the SKEMPI database (Jankauskait ˙ e et al., 2019) as a binary classification task.”
Multi-Scale Representation Learning on Proteins (2022, methodology): “…2018, Rao et al., 2019, Bepler and Berger, 2019, Elnaggar et al., 2020] as well as methods based on the three-dimensional structure of proteins [Townshend et al., 2020, Hermosilla et al., 2021], and recent methods using geometric deep learning on protein molecular surfaces [Gainza et al.,…”
A guide to machine learning for biologists (2021, background): “For the field to progress, it will be necessary to develop benchmark datasets and validation tasks 116 , such as ProteinNet 117 , ATOM3D 118 and TAPE 119 , and for these to become widely used.”
Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs (2022, background): “There are two main ways to represent atomistic graphs (Townshend et al., 2021), which are chemical bond graphs, sometimes denoted as 2D graphs, and 3D spatial graphs.”
TorchDrug: A Powerful and Flexible Machine Learning Platform for Drug Discovery (2022, methodology): “Other implementations like Therapeutics Data Commons (TDC) (Huang et al., 2021) and ATOM3D (Townshend et al., 2020) focus on datasets and evaluation toolkits for drug discovery tasks, but lack implementation of models.”
Equivariant Graph Neural Networks for 3D Macromolecular Structure (2021, unknown): “The tasks in ATOM3D span several orders of magnitude in both dataset size and structure size (a rough proxy for task complexity), and therefore range from data-rich (SMP, RES) to data-poor (MSP, LEP).”
Scientific discovery in the age of artificial intelligence (2023, background): “Standardized and transparent formats are required to alleviate the workload of data processing 159,192–196 .”
Efficient and Accurate Physics-aware Multiplex Graph Neural Networks for 3D Small Molecules and Macromolecule Complexes (2022, background): “However, they do not take account of the 3D structures of molecules, which are critical for molecular representations [36].”