3D deep convolutional neural networks for amino acid environment similarity analysis

By Wen Torng, Russ B. Altman. 2017

147 citations recorded by the source corpus.

Central to protein biology is the understanding of how structural elements give rise to observed function. The surfeit of protein structural data enables development of computational methods to systematically derive rules governing structural-functional relationships. However, performance of these methods depends critically on the choice of protein structural representation. Most current methods rely on features that are manually selected based on knowledge about protein structures. These are often general-purpose but not optimized for the specific application of interest. In this paper, we present a general framework that applies 3D convolutional neural network (3DCNN) technology to structure-based protein analysis. The framework automatically extracts task-specific features from the raw atom distribution, driven by supervised labels. As a pilot study, we use our network to analyze local protein microenvironments surrounding the 20 amino acids, and predict the amino acids most compatible with environments within a protein structure. To further validate the power of our method, we construct two amino acid substitution matrices from the prediction statistics and use them to predict effects of mutations in T4 lysozyme structures. Our deep 3DCNN achieves a two-fold increase in prediction accuracy compared to models that employ conventional hand-engineered features and successfully recapitulates known information about similar and different microenvironments. Models built from our predictions and substitution matrices achieve an 85% accuracy predicting outcomes of the T4 lysozyme mutation variants. Our substitution matrices contain rich information relevant to mutation analysis compared to well-established substitution matrices. Finally, we present a visualization method to inspect the individual contributions of each atom to the classification decisions. End-to-end trained deep learning networks consistently outperform methods using hand-engineered features, suggesting that the 3DCNN framework is well suited for analysis of protein microenvironments and may be useful for other protein structural analyses.

Subfields: Materials Chemistry, Molecular Biology

How researchers see this paper

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

In Materials Chemistry

In Molecular Biology

Citations in the corpus

Tensor Field Networks: Rotation- and Translation-Equivariant Neural Networks for 3D Point Clouds (2018, neutral): “The other major approach to modeling 3D atomic systems is to voxelize the space [16–18].”
Hierarchical, rotation‐equivariant neural networks to select structural models of protein complexes (2020, unknown): “Three-dimensional convolutional neural networks (3D CNNs) have been applied to several problems in structural biology, including the prediction of protein structure, protein-protein interfaces, and protein-ligand binding.(23-29) Although powerful, conventional 3D CNNs are suboptimal in two ways when”
High precision protein functional site detection using 3D convolutional neural networks (2018, methodology): “We have previously applied 3D convolutional neural networks (3DCNNs) to amino acid similarity analysis and showed deep learning framework outperformed conventional feature-based algorithms (Torng and Altman, 2017).”
End-to-End Learning on 3D Protein Structure for Interface Prediction (2018, methodology): “3D convolutional networks have been used for protein-ligand binding affinity [14], [15], [16], as well as for protein fold quality assessment [17], protein structure classication [18], fingerprint prediction [19], and filling in missing amino acids [20].”
Spherical convolutions and their application in molecular modelling (2017, unknown): “Finally, we note that while this manuscript was in review, another paper on the application of convolutional neural networks for predicting amino acid preferences conditioned on structural environments was published, by Torng and Altman (Torng and Altman, 2017).”
A structure-based deep learning framework for protein engineering (2019, methodology): “1b) using the original training and testing sets (32,760 and 1601 structures, respectively) 9 .”
Deep convolutional networks for quality assessment of protein folds (2018, methodology): “More in line with the “end-to-end” spirit of deep learning, methods using as input a 3D representation of the structure have been developed to score protein-ligand poses [16], [17], to predict ligand-binding protein pockets [18], and to predict the effect of a protein mutation [19].”
Protein sequence design with a learned potential (2020, methodology): “Previous machine learning models for this particular task have only been applied to architecture class, secondary structure, and ∆∆G prediction [66, 67, 68].”
Predicting the locations of cryptic pockets from single protein structures using the PocketMiner graph neural network (2022, methodology): “For training our baseline 3D CNN model, we use the featurization described in Torng et al. 2017 75 , which creates a 3D input with four different channels, with one channel per atom type for: carbon, oxygen, sulfur, and nitrogen atoms.”
ATOM3D: Tasks On Molecules in Three Dimensions (2020, methodology): “We can understand this role by predicting the propensity for different amino acids at a given protein site based on the surrounding structural environment [Torng and Altman, 2017].”