CoChat
Sign Up
Cited by 142Protein Structure and DynamicsMachine Learning in BioinformaticsComputational Drug Discovery Methods

The DOI matches its Crossref record; 84 of the 84 references it cites were found in a registry or an index.

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

Open access
Torng, Wen, Altman, Russ B.·2017·BMC Bioinformatics
DOI 10.1186/s12859-017-1702-0PMC PMC5472009
Open this paper in CoChat — free accountOpen PDF

Abstract

BACKGROUND: 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. RESULTS: 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. CONCLUSIONS: 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.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1186/s12859-017-1702-0

Cites

84 of 84 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Local box sampling and extraction. a For each structure in the training and test structure sets, we placed a 3D grid with 10 Å spacing to sample positions in the protein for local box extraction. The teal spheres represent the sampled grid positions. (For illustration purpose, a grid size of 25 Å instead of 10 Å is shown here). b For each sampled position, the nearest atom (pink sphere) to the sampled position (teal sphere) is first identified. c The amino acid which this atom belongs to is then assigned as the central amino acid. The selected amino acids are highlighted in red and the atoms are shown as dotted spheres. d A local box of 20 Å is then defined around the central amino acid, centering on the Cβ. For each amino acid microenvironment, a local box is extracted around the amino acid using the following procedure: e Backbone atoms of the center amino acid is first used to calculate the orthogonal axes for box extraction. f A 20 Å box is extracted around the Cβ atom of the center amino acid using the defined orientation. g Side-chain atoms of the center amino acid are removed. The extracted box is then labeled with the removed amino acid side-chain type
Fig. 1 Local box sampling and extraction. a For each structure in the training and test structure sets, we placed a 3D grid with 10 Å spacing to sample positions in the protein for local box extraction. The teal spheres represent the sampled grid positions. (For illustration purpose, a grid size of 25 Å instead of 10 Å is shown here). b For each sampled position, the nearest atom (pink sphere) to the sampled position (teal sphere) is first identified. c The amino acid which this atom belongs to is then assigned as the central amino acid. The selected amino acids are highlighted in red and the atoms are shown as dotted spheres. d A local box of 20 Å is then defined around the central amino acid, centering on the Cβ. For each amino acid microenvironment, a local box is extracted around the amino acid using the following procedure: e Backbone atoms of the center amino acid is first used to calculate the orthogonal axes for box extraction. f A 20 Å box is extracted around the Cβ atom of the center amino acid using the defined orientation. g Side-chain atoms of the center amino acid are removed. The extracted box is then labeled with the removed amino acid side-chain typeCC BY · PubMed Central
Fig. 2 Local box featurization. a Local structure in each 20 Å box is first decomposed into Oxygen, Carbon, Nitrogen, and Sulfur channels. b Each atom type channel structure is divided into 3D 1-Å voxels, within which the presence of atom of the corresponding atom type is recorded. Within each channel, Gaussian filters are applied to the discrete counts to approximate the atom connectivity and electron delocalization. c The resulting numerical 3D matrices of each atom type channel are then stacked together as different input channels, resulting in a (4, 20, 20, 20) 4D–tensor, which will serve as an input example to our 3DCNN
Fig. 2 Local box featurization. a Local structure in each 20 Å box is first decomposed into Oxygen, Carbon, Nitrogen, and Sulfur channels. b Each atom type channel structure is divided into 3D 1-Å voxels, within which the presence of atom of the corresponding atom type is recorded. Within each channel, Gaussian filters are applied to the discrete counts to approximate the atom connectivity and electron delocalization. c The resulting numerical 3D matrices of each atom type channel are then stacked together as different input channels, resulting in a (4, 20, 20, 20) 4D–tensor, which will serve as an input example to our 3DCNNCC BY · PubMed Central
Fig. 3 The FEATURE program. FEATURE captures the physicochemical information around a point of interest in protein structure by segmenting the local environment into six concentric shells, each of 1.25 Å in thickness. Within each shell, FEATURE evaluates 80 physicochemical properties including atom type, residue class, hydrophobicity, and secondary structure. This enables conversion of a local structural environment into a numeric vector of length 480
Fig. 3 The FEATURE program. FEATURE captures the physicochemical information around a point of interest in protein structure by segmenting the local environment into six concentric shells, each of 1.25 Å in thickness. Within each shell, FEATURE evaluates 80 physicochemical properties including atom type, residue class, hydrophobicity, and secondary structure. This enables conversion of a local structural environment into a numeric vector of length 480CC BY · PubMed Central
Fig. 4 Schematic diagram of the Deep 3D Convolutional Neural Network and FEATURE-Softmax Classifier models. a Deep 3D Convolutional Neural Network. The feature extraction stage includes 3D convolutional and max-pooling layers. 3D filters in the 3D convolutional layers search for recurrent spatial patterns that best capture the local biochemical features to separate the 20 amino acid microenvironments. Max Pooling layers perform down-sampling to the input to increase translational invariances of the network. By following the 3DCNN and 3D Max-Pooling layers with fully connected layers, the pooled filter responses of all filters across all positions in the protein box can be integrated. The integrated information is then fed to the Softmax classifier layer to calculate class probabilities and to make the final predictions. Prediction error drives parameter updates of the trainable parameters in the classifier, fully connected layers, and convolutional filters to learn the best feature for the optimal performances. b The FEATURE Softmax Classifier. The FEATURE Softmax model begins with an input layer, which takes in FEATURE vectors, followed by two fully-connected layers, and ends with a Softmax classifier layer. In this case, the input layer is equivalent to the feature extraction stage. In contrast to 3DCNN, the prediction error only drives parameter learning of the fully connected layers and classifier. The input feature is fixed during the whole training process
Fig. 4 Schematic diagram of the Deep 3D Convolutional Neural Network and FEATURE-Softmax Classifier models. a Deep 3D Convolutional Neural Network. The feature extraction stage includes 3D convolutional and max-pooling layers. 3D filters in the 3D convolutional layers search for recurrent spatial patterns that best capture the local biochemical features to separate the 20 amino acid microenvironments. Max Pooling layers perform down-sampling to the input to increase translational invariances of the network. By following the 3DCNN and 3D Max-Pooling layers with fully connected layers, the pooled filter responses of all filters across all positions in the protein box can be integrated. The integrated information is then fed to the Softmax classifier layer to calculate class probabilities and to make the final predictions. Prediction error drives parameter updates of the trainable parameters in the classifier, fully connected layers, and convolutional filters to learn the best feature for the optimal performances. b The FEATURE Softmax Classifier. The FEATURE Softmax model begins with an input layer, which takes in FEATURE vectors, followed by two fully-connected layers, and ends with a Softmax classifier layer. In this case, the input layer is equivalent to the feature extraction stage. In contrast to 3DCNN, the prediction error only drives parameter learning of the fully connected layers and classifier. The input feature is fixed during the whole training processCC BY · PubMed Central
Fig. 5 Confusion matrices for predictions of the 20 amino acid microenvironments. Predictions on the training and test datasets using 3DCNN and FEATURE Softmax Classifier are summarized into confusion matrices to inspect the propensity of each microenvironment type to be predicted as one another. The 20 amino acids are arranged according to knowledge-based amino acid groups, where amino acids known to be biochemically similar are adjacent. The ith, jth element of the matrices shows the probability of examples of true label i being predicted as label j. The probability is represented in heat map colors. a 3DCNN-Train. b 3DCNN-Test. Local block structures in the confusion matrices for 3DCNN demonstrate similarities and differences between amino acid microenvironments. For example, phenylalanine (F), tryptophan (W), and tyrosine (Y) form a hydrophobic and aromatic block. Similar block structure is observed for test predictions. The captured features are robust across protein families. c FEATURE-Train. d FEATURE-Test. Block structures are less evident in the confusion matrices for the FEATURE Softmax classifier
Fig. 5 Confusion matrices for predictions of the 20 amino acid microenvironments. Predictions on the training and test datasets using 3DCNN and FEATURE Softmax Classifier are summarized into confusion matrices to inspect the propensity of each microenvironment type to be predicted as one another. The 20 amino acids are arranged according to knowledge-based amino acid groups, where amino acids known to be biochemically similar are adjacent. The ith, jth element of the matrices shows the probability of examples of true label i being predicted as label j. The probability is represented in heat map colors. a 3DCNN-Train. b 3DCNN-Test. Local block structures in the confusion matrices for 3DCNN demonstrate similarities and differences between amino acid microenvironments. For example, phenylalanine (F), tryptophan (W), and tyrosine (Y) form a hydrophobic and aromatic block. Similar block structure is observed for test predictions. The captured features are robust across protein families. c FEATURE-Train. d FEATURE-Test. Block structures are less evident in the confusion matrices for the FEATURE Softmax classifierCC BY · PubMed Central
Fig. 6 Hierarchical Clustering of normalized confusion matrices. The ith, jth element of the row-normalized matrices shows the probability of examples of true label i being predicted as label j. The probability is represented in heat map colors. Hierarchical clustering reveals similarities between amino acid microenvironments in terms of their propensities to be assigned to the 20 amino acid types. a 3DCNN-Training. Amino acid groupings discovered by our 3DCNN generally agree with known amino acid similarities. Six clusters were discovered by our network. The first cluster includes phenylalanine, tryptophan, and tyrosine. These are the three amino acids known to be hydrophobic and aromatic. The second and third clusters comprises valine, isoleucine and leucine, methionine respectively, which are all non-polar and aliphatic. The polar amino acids form the fourth cluster. Amino acids with known distinct properties, glycine and cysteine do not form local blocks with the other amino acids. b 3DCNN-Test Groupings generated for the test examples are consistent with the training counterparts. c FEATURE-Softmax-Training. d FEATURE-Softmax-Test Clustering on the FEATURE Softmax classifier generates much coarser amino acid groupings than the ones discovered by 3DCNN. The two major groups, hydrophobic and polar amino acids are separated
Fig. 6 Hierarchical Clustering of normalized confusion matrices. The ith, jth element of the row-normalized matrices shows the probability of examples of true label i being predicted as label j. The probability is represented in heat map colors. Hierarchical clustering reveals similarities between amino acid microenvironments in terms of their propensities to be assigned to the 20 amino acid types. a 3DCNN-Training. Amino acid groupings discovered by our 3DCNN generally agree with known amino acid similarities. Six clusters were discovered by our network. The first cluster includes phenylalanine, tryptophan, and tyrosine. These are the three amino acids known to be hydrophobic and aromatic. The second and third clusters comprises valine, isoleucine and leucine, methionine respectively, which are all non-polar and aliphatic. The polar amino acids form the fourth cluster. Amino acids with known distinct properties, glycine and cysteine do not form local blocks with the other amino acids. b 3DCNN-Test Groupings generated for the test examples are consistent with the training counterparts. c FEATURE-Softmax-Training. d FEATURE-Softmax-Test Clustering on the FEATURE Softmax classifier generates much coarser amino acid groupings than the ones discovered by 3DCNN. The two major groups, hydrophobic and polar amino acids are separatedCC BY · PubMed Central
Fig. 7 Scatter plots of similarity scores between S freq, S dot and benchmarking matrices BLOSUM62, PAM250, WAC. a S freq- BLOSUM62. b S freq- PAM250. c S freq- WAC. d S dot- BLOSUM62. e S dot- PAM250. f S dot-WAC. g WAC- BLOSUM62. h WAC- PAM250. i PAM250- BLOSUM62. S freq shows generally good correlations with BLOSUM62 and PAM250; S dot shows strong correlation with BLOSUM62 and PAM250; S freq and S dot show no significant correlations with WAC
Fig. 7 Scatter plots of similarity scores between S freq, S dot and benchmarking matrices BLOSUM62, PAM250, WAC. a S freq- BLOSUM62. b S freq- PAM250. c S freq- WAC. d S dot- BLOSUM62. e S dot- PAM250. f S dot-WAC. g WAC- BLOSUM62. h WAC- PAM250. i PAM250- BLOSUM62. S freq shows generally good correlations with BLOSUM62 and PAM250; S dot shows strong correlation with BLOSUM62 and PAM250; S freq and S dot show no significant correlations with WACCC BY · PubMed Central
Fig. 8 Importance visualization of local amino acid microenvironments. Visualizations of importance scores of each input atom are displayed as heat maps. The color demonstrates how each atom within the local box contribute to the decision. The importance scores range from 0 to 100. Atoms with the lowest important scores (<20) are shown in white, and the red to blue heat map spectrum highlights the most important to the least important atoms. Transparent light pink spheres are drawn to show the space previously occupied by the removed residue. a Microenvironment surrounding a key ASP residue at the EF_HAND calcium binding site (PDB: 1A2X, ASP 63.) Our importance score map indicates that the correct prediction relies on the two nitrogen atoms, which are in close proximity to the electronegative oxygen atoms of the removed aspartic acid residue. b Microenvironment surrounding a PROSITE INSULIN motif, with the key CYS residue removed (PDB:1IZA, CYS 7.) The 3DCNN network made the correct prediction of CYS primarily based on the SG atom from a nearby cysteine residue. This SG atom originally forms disulfide bond with the SG atom of the removed cysteine residue. The unique disulfide bond pattern was implicitly captured by our network to facilitate the classification. c Microenvironment surrounding a phenylalanine residue (PDB: 1ZPL, PHE_52). The three highlighted regions in the heat map are the side-chain atoms of VAL 54, ILE 48, VAL 2, all of which comprise of non-polar carbon atoms. d Microenvironment surrounding a valine residue (PDB: 1VJ9 [64], VAL 200). The highlighted atom groups are the side-chains of MET 207, TRP 29 and ILE 121, which are similarly non-polar
Fig. 8 Importance visualization of local amino acid microenvironments. Visualizations of importance scores of each input atom are displayed as heat maps. The color demonstrates how each atom within the local box contribute to the decision. The importance scores range from 0 to 100. Atoms with the lowest important scores (<20) are shown in white, and the red to blue heat map spectrum highlights the most important to the least important atoms. Transparent light pink spheres are drawn to show the space previously occupied by the removed residue. a Microenvironment surrounding a key ASP residue at the EF_HAND calcium binding site (PDB: 1A2X, ASP 63.) Our importance score map indicates that the correct prediction relies on the two nitrogen atoms, which are in close proximity to the electronegative oxygen atoms of the removed aspartic acid residue. b Microenvironment surrounding a PROSITE INSULIN motif, with the key CYS residue removed (PDB:1IZA, CYS 7.) The 3DCNN network made the correct prediction of CYS primarily based on the SG atom from a nearby cysteine residue. This SG atom originally forms disulfide bond with the SG atom of the removed cysteine residue. The unique disulfide bond pattern was implicitly captured by our network to facilitate the classification. c Microenvironment surrounding a phenylalanine residue (PDB: 1ZPL, PHE_52). The three highlighted regions in the heat map are the side-chain atoms of VAL 54, ILE 48, VAL 2, all of which comprise of non-polar carbon atoms. d Microenvironment surrounding a valine residue (PDB: 1VJ9 [64], VAL 200). The highlighted atom groups are the side-chains of MET 207, TRP 29 and ILE 121, which are similarly non-polarCC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 82 references

Research works printed but not linked: 2

  1. C. M. Bishop, Pattern recognition. 2006.
  2. B. C. Dayhoff, M.O., Schwartz, R. and Orcutt, “A Model of Evolutionary Change in Proteins,” in Atlas of protein sequence and structure, Volume 5,., National Biomedical Research Foundation Silver Spring, MD, 1978, pp. 345–358

Cited by

Sorted by citations

See all 142

Related

Metadata

Authors
Torng, W., & Altman, R. B.
Affiliations
Stanford University (US)
Corresponding author
Altman, Russ B.
ORCID
Venue
BMC Bioinformatics
Publisher
Springer Science and Business Media LLC
Published
2017-06-14
Volume
18
Issue
1
Pages
302-302
ISSN
1471-2105 (ISSN-L)
DOI
10.1186/s12859-017-1702-0
PMC
PMC5472009
OpenAlex
W2625768267
Cited by
142
References
82
Citations by year
2026: 6 · 2025: 13 · 2024: 12 · 2023: 23 · 2022: 21 · 2021: 18 · 2020: 15 · 2019: 20 · 2018: 14
FWCI
5.48
Type
journal article
Language
en
Open access
Licence
Locations
  • BMC Bioinformatics (publishedVersion, cc-by)
  • Europe PMC (PubMed Central) (publishedVersion, other-oa)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
National Institutes of Health (GM102365, LM05652, HL117798)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Machine Learning in Bioinformatics
  • Life Sciences › Pharmacology, Toxicology and Pharmaceutics › Drug Discovery › Computational Drug Discovery Methods
Keywords
3D convolutional neural networks, protein microenvironment, protein structure representation, amino acid substitution matrices, T4 lysozyme, mutation prediction, structure-function relationship, end-to-end deep learning, feature visualization
MeSH terms
Amino Acids, Animals, Brain, Nerve Net, ROC Curve, In Situ Hybridization, Area Under Curve, Embryonic Development, Mice, Transcriptome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review