Welcome to Bio3D-web
- Online analysis of user defined protein structure ensembles -
Overview
Bio3D-web is a new online application for the user friendly investigation of protein structure ensembles.
Major functionality allows you to map and explore the structural, conformational and internal dynamic properties of proteins for which there are high resolution structures available. Read More>>
Example Applications
Live Demo's
Frequently Asked Questions
1. What is Bio3D-web and how does it differ from the Bio3D package?
Bio3D-web is a new online web server for the user friendly exploratory analysis of protein sequence-structure-dynamics relationship. Bio3D-web is powered by the well established Bio3D R package for structural bioinformatics (Grant et al. (2006)).
Bio3D-web does not require any installation or programming skills – you explore through an easy to use online interface. This is in contrast to the conventional Bio3D package, which typically requires installation on your own hardware, knowledge of the R-Bio3D language, and use of the unix like command line.
2. Is this just another online NMA server?
No. There are lots of excellent normal mode analysis servers already out there such as WebNM@, elNémo, AD-ENM, ANM and iMODS. These offer NMA calculation capabilities for individual biomolecular structures.
Bio3D-web is unique in providing more expansive and integrated functionality for the identification, comparison and detailed analysis of large homologous structure sets online. We designed Bio3D-web to increase the accessibility and decrease the entry barrier to performing advanced comparative sequence, structure and dynamics analysis across user defined structure sets not just single structures.
3. Why not just use Bio3D proper or ProDy?
Both Bio3D and ProDy (a related python library developed by others) offer much more functionality than is provided in this online application. If you can use these packages productively then we encourage you to do so. One of our main motivations for developing Bio3D-web was to allow new users to be productive with methods like PCA and eNMA without having to first learn Bio3D usage. We hope you will find Bio3D-web useful and then feel motivated to use the conventional Bio3D R-package on your own computers and for your own custom analysis.
4. What is PCA and why is it useful for analysis of biomolecular structures?
Principal component analysis (PCA) is a well established statistical method that is most commonly used as a dimensionally reduction technique for multivariate data analysis - that is input data that has many dimensions, e.g. many different atomic coordinates that have been measured for multiple experimental structures.
Essentially, the PCA performed here aims to succinctly ‘map out’ the conformational relationships in large sets of protein structures and provide quantitate insight into the structural regions that contribute to any distinct conformations identified.
More explicitly, Bio3D-web utilizes PCA to provide a new condensed view of large structural datasets. This condensed view is basically a re-framing that retains the essential essence of the entire coordinate data. The new view is given in terms of what are known as principal components. These principal components are new directions in the data along which there is maximal variance - or more simply put, the directions where our set of structures differ most (i.e. are most spread out). The whole idea of PCA is to find these new directions of maximal variation in the coordinate data and use them to better understand major conformational features of the dataset.
The output of PCA includes a lower dimensional ‘reframing’ of your complete structure set that can simplify visualization and make it easier to uncover and further analyze interesting underlying structure relationships. These relationships can be hard to see in the original coordinate data (e.g. from just looking at superposed structures) because you might have many original dimensions to examine. PCA is also particularly useful as it allows you to qualitatively assess which regions of your structures are contributing most to the revealed structure relationships.
We encourage users that are not familiar with PCA to look into many of the great resources available online (e.g. the Bio3D website).
5. What is eNMA and why is it useful for analysis of biomolecular structures?
Normal modes analysis (NMA) is a computational technique to characterize all possible deformations a protein can undergo. These motions are conveniently sorted with respect to the energy needed to deform the protein along the particular normal mode vector. This computational technique is particularly suited to probe large-scale collective motions typically associated with protein function.
NMA application most often involves the analysis of only a single protein structure. As the normal modes are sensitive to the specific protein conformation for which they are calculated, the exclusion of alternative protein conformations provides only a limited picture of the overall flexibility of the protein under different conditions.
A more complete picture of protein flexibility can be obtained by performing NMA across all structure in an ensemble in a way that facilities the interoperation of structural similarity and dissimilarity trends. This allows a user to explore dynamic trends of all crystalized states in relation to each other without the conventional caveat of potentially over-interpreting the differences between extreme cases or a single artifactual structure. Furthermore, by carefully contrasting the fluctuation profiles one can provide new information on state specific global and local dynamics of potential functional relevance.
6. What are the major limitations of the current version of Bio3D-web?
A major current limitation is our restriction to analyzing only single chains from multi-chain PDB structures. Future versions of Bio3D-web will include the ability to perform single-chain, multi-chain and reconstructed biounit analysis. For now, if you would like to perform this type of analysis you should use the full Bio3D R package and the new biounit() function.
Another limitation is the comparatively slow performance of the ensemble normal mode analysis tab. Note that due to available hardware limitations we currently perform eNMA in series and thus restrict the total number of structures analyzed (even though our underlying code and approach is now paralyzed). Bio3D-web currently runs entirely on a small virtual machine. We plan to improve performance of the eNMA tab to seconds even for many hundreds of structures by linking to suitable cluster computing resources. Please contact us if this is something you would like to do now.
Our 3D structure viewer has limited interactivity and does not render trajectories as movies etc. For example, you can not click on a region of structure and find out what residues are involved in a large-scale motion. We are currently using the inbuilt Bio3D view.pdbs() function via WebGL but are exploring using PV for future versions. In this regard please note that we provide links/buttons to view your superposed structure ensembles, PCA and eNMA results in PyMOL on your own computer. We also provide PDB file download options that should allow you to more comprehensively view your results in other powerful molecular viewers including VMD and Chimera .
7. How is the search for related protein structures performed?
The search is sequence based using pHMMER over the PDB database.
8. What is the 'BitScore of Alignment to Input Structure'?
The bitscore is a score that describes the overall quality of the alignment between the query sequence and the search result. High bitscore corresponds to high sequence similarity. We report the bit scores from the HMMER search.
9. How is the invariant 'structural core' identified?
The algorithm iteratively refines an initial structural superposition determined from a multiple alignment. This involves iterated rounds of superposition, where at each round the position(s) displaying the largest differences is(are) excluded from the dataset (Grant et al. (2006)).
10. Is the source code available and can I run Bio3D-web on my own hardware'?
The complete Bio3D-web source code, like the underlying Bio3D package itself, is made fully available under a GPL2 license. Instructions for running on any computer running R are available here.
11. How are the principal components calculated?
The principal components is calcualted from the superimposed coordinates excluding the gap containing columns (Skjaerven et al. (2014)).
12. How are the normal modes calculated?
The normal modes for each structure in the representative ensemble are calculated according to Fuglebakk et al. (2012) (see also Skjaerven et al. (2014)). The C-alpha force field developed by Konrad Hinsen (Hinsen et al. (2000)) is used for normal mode calculation. For a discussion and comparison of other force-fields and NMA methods see Yao et al. (2016)
13. How to cite Bio3D-web?
If you have used Bio3D-web, please consider citing the following reference that describes this work:
- Skjaerven et al. Online interactive analysis of protein structure ensembles with Bio3D-web (2016) Bioinformatics 32(22), 3510—3512. doi:10.1093/bioinformatics/btw482
- Grant et al. Bio3D: An R package for the comparative analysis of protein structures (2006) Bioinformatics 22, 2695-2696
- Skjaerven et al. Integrating protein structural dynamics and evolutionary analysis with Bio3D. (2014) BMC Bioinformatics 15, 399
14. How to contact us with your questions and suggestions?
Your questions and comments are important to us. If you like or do not like what we are doing, then please get in touch. Also, if there are new features that you would like to have added to the server, then drop us a line and we will see what we can do!
For the moment please use our BitBucket based issues tracker for any questions regarding the use of Bio3D-web or the larger Bio3D software package.
To expedite our response to your questions, please provide us with as much information as possible so that we can recreate the problem. Useful things to include are:
- State if it is a problem with the Bio3D-web server or a local software installation.
- Input data sufficient to recreate the problem.
- Details of which web browser and the version?
Many questions are also covered in the Bio3D documentation, so it may be worth browsing them before posting an issue.
Structure Search
A) Input Structure(s) or Sequence
Please enter either a single PDB code of interest, a single protein sequence, or multiple related PDB codes (see the Help page for more details).Multiple PDB Summary
Annotation from PFAM of requested multiple PDB structures.Protein Sequence Summary
Structure Summary
Input Structure Visualization
Note: Use your mouse to drag, rotate and zoom the structureInput PDB Read Log
Overview
This site provides an online interface to several Bio3D tools for comparative protein structure analysis.
Methods include (1) Searching for related structures, (2) Alignment of selected structures, (3) Fitting based on rigid core positions, (4) PCA (principal component analysis) for inter-conformer characterization and (5) eNMA (ensemble normal mode analysis) for additional structure dynamic characterization.
Each of these analysis steps is implemented as consecutive tabs accessible from the top navigation bar.
Start your analysis by entering a PDB code of interest and then proceed by navigating through the above tabs or following the NEXT buttons.
For a detailed usage guide please download and consult the Bio3D-web tutorial [ PDF ].
B) Hit selection for further analysis
Optional filtering and refinement (via similarity threshold specification) of related structures for further analysis.BLAST options not available for multiple PDB input
C) Optional filtering of related structures for further analysis
Optionally select (or de-select) structures by clicking to highlight their entries in the below table. This allows for finer grained selection than the sliders in panel B above.Structure chain annotation
Input structures may consist of multiple chains. If not explicitly specified in panel A above (via using underscore chain in the input identifier e.g. '1KJY_A') all of these chains will be included below and be subject to further analysis.Sequence Alignment
Sequence Alignment and Analysis
In this tab all PDB structures selected in the previous (SEARCH) tab are parsed and their sequences aligned. The tab displays the final alignment of the selected PDB structures, as well as basic analysis of sequence identity.
You can optionally exclude structures, either manually (see 'Exlcude / include hits'), or automatically omit structures with missing in-structure residues. Functionality for uploading a corrected / revised sequence alignment (FASTA format) is also provided.
The sequence alignment is performed with MUSCLE and is displayed in blocks of up to 80 columns with a row per PDB structure. Conserved columns are annotated with an asterisk (*) while columns containing similar amino acids are depicted with a hat (^). Detailed information (including PDB residue number, and position in alignment) on each residue in the alignment can be obtained by hovering over the the residue.
Basic analyses of sequence identity entails clustering analysis, as well as entropy and sequence conservation. These analyses are available as plot dendrograms, heatmap and histogram plot.
A) Alignment summary
Edit alignment
Individual structures can be filtered (i.e. removed) from the alignment by deleting their identifier from the list below. Alternatively, the alignment itself can be downloaded, edited and re-loaded via the Upload option.Download FASTA alignment file
B) Sequence alignment analysis
Download Figures and Data
Conservation (PDF)D) Optional sequence alignment display
Rendering the alignment for display can be time consuming for large data sets (e.g. > 50 PDB structures of substantial sequence length).Structure Superposition
Structure superposition
In the FIT tab all structures are superimposed based on the sequence alignment obtained in the ALIGN tab. By default, the structures are superimposed on the identified invariant core. Alternatively, a conventional superposition based on all protein residues can be performed.
Basic structural analysis in this tab entails the calculation of all pair-wise structural deviations (RMSD), fluctuation analysis (RMSF), and structure visualization. Based on the pair-wise RMSD values the structures can be clustered into groups of similar conformations. The results of the RMSD and clustering analyses are available as a heatmap, clustering dendrogram, and RMSD histogram.
The algorithm for identifying the invariant core attempts to iteratively refine the initial structural superposition determined from the multiple alignment. This involves iterated rounds of superposition, where at each round the position(s) displaying the largest differences is(are) excluded from the dataset. The identified core can be visualized either in the browser, or in PyMOL by downloading the PyMOL state (.pse) file. A summary of the core can be found in the bottom of the page.
A) Superposed PDB viewing options
Download options:
Aligned PDBs PyMOL SessionFilter/toggle displayed PDBs
B) Initial structure analysis
C) Residue fluctuations
D) Structural core and RMSD details
Summary of invariant core
RMSD summary
Cluster representatives
Principal Component Analysis
Principal Component Analysis
In the PCA tab, a principal component analysis (PCA) is performed based on the coordinates of the superimposed structures in the FIT tab. PCA is an effective approach to capture and characterize inter-conformer relationships. The motions described by the principal components (PCs) can be visualized either in the browser, or by downloading a PDB trajectory file or a PyMOL session (.pse) containing a vector field representation.
The figure above shows the first principal component obtained from the 32 PDB structures of the enzyme family aromatic amino acid hydroxylases. The analysis reveals the presence of a sub-domain within the catalytic domain which obtains a more compact conformation upon substrate binding.
The figure above shows the conformer plot of 90 E. coli DHFR structures. The PCA reveals that the ensemble can be divided into three major groups along their first two principal components. These conformers display either a closed, occluded,oranopenconformationof two active site loops. The conformer plot displays a low-dimensional representation of conformational variability described by two selected PCs (typically PC-1 and PC-2).
A) Principal component visualization
Download options:
PDB Trajectory PyMOL SessionB) Conformer analysis
Label options
PCA conformer plot annotation
Highlight structures in conformer plot by clicking their entries in the below table (only for plot type '2D Scatter').C) Residue contributions
Download PDF Figures
Residue contributions (PDF)Ensemble Normal Mode Analysis
Ensemble Normal Mode Analysis
Normal mode analysis (NMA) characterizes protein dynamics proximal to a local energy minimum. This implies that predictions from NMA approaches can be dependent on the (often arbitrarily chosen) initial structure used for calculations. NMA applied to a single structure can thus be strongly biased to the specific state represented by the structure and may fail to capture the more global picture of structural dynamics (Tama and Sanejouand, 2001). This input structure dependency of NMA also provides an effective way to characterize and potentially classify protein structures via a comparison of NMA derived dynamics. Using this approach additional structural dynamic insight can be obtained that is not readily available from conventional structural comparison. This realization led to our development of the ensemble NMA approach (eNMA), which calculates modes for all structures in a systematic and unified manner.
Critically, the NMA here is performed on all structures in a way that facilities the interpretation of their structural similarity and dissimilarity trends (Yao et al., 2016). This facilitates the quantitative comparison of normal modes across heterogeneous structures (with unequal sequence composition and length). This also allows a user to more fully explore dynamic trends in selected crystallized states in relation to each other without the conventional caveat of potentially over-interpreting the differences between extreme cases or a single artifactual structure.
Clustering based on calculated eNMA modes and characterization of each cluster (or state) can effectively reveal the dynamical heterogeneity that is missing in single structure based NMA. Indeed, detailed comparison shows that ensemble NMA has an overall improvement over single structure NMA in predicting many important structural dynamic properties (Yao et al., 2016).
Furthermore, eNMA can directly reveal state specific trends and differences in residue wise fluctuations of potential functional relevance.
The actual motions described by each normal mode can be visualized either in the browser, or by downloading a PDB trajectory file or a PyMOL session (.pse) containing a vector field representation.
A) Filter structures
Focus the structure ensemble by filtering out similar structures to reduce the computational load for NMA. Interactively (de)select structures by clicking the PC conformer plot.D) Compare PCA and NMA
The NMs of the selected 'reference' PDB are compared to the PCs derived from the ensemble of PDB structures. Note: Only the filtered structures are used in the comparison (panel A).E) Overlap analysis
The NMs of the selected 'reference' PDB are compared to the vector describing the difference in conformation between two selected structures (i.e. the reference structure and each of the selected PDBs from the table below). Overlap (PDF)Cluster heatmaps
Row side colors correspond to RMSIP clustering for all plots below. Column side colors correspond to clustering from NMA, FIT, and PCA tab, respectively.RMSIP heatmap
RMSD heatmap
PCA heatmap
Download Options
Select a format for your summary report and click the Download button to save to your computer.Save or load your analysis
I want to:
Save your analysis
The results from your analyses will be saved on the server for 30 days.
Load your analysis
You can revisit your data by pasting the session ID in the box below.