Home
FFFold is a web application providing a simple-to-use interface for the local optimisation of protein structures predicted by the AlphaFold2 algorithm [Jumper2021] and deposited in the AlphaFold DB database [Varadi2022]. FFFold optimizes only residues predicted with confidence less than 90. The application consists of three pages: Main Page, the Optimisation progress page, and the Optimisation results page.
The main page allows uploading the chosen structure from AlphaFold DB and set the pH at which the optimised structure is to be protonated. The structure must be defined by its UniProt [UniProt2019] accession number (so-called UniProt ID or UniProt code):
The UniProt codes of the proteins can be obtained directly from AlphaFold DB. Here, you can search for your protein, gene, or organism. Examples of UniProt codes, which are accepted by FFFold: A0A159JYF7, Q96247, F4HT52, or C0SV66. Using UniProt entry name (e.g. A0A159JYF7_9DIPT, AUX1_ARATH, F4HT52_ARATH) or UniProt accession number with identification of fragments (e.g. Q8WZ42-F1, Q8WZ42-F2) is not supported by FFFold. An alternative option is AlpfaFold DB Identifier (e.g. AF-L8BU87-F1). Moreover, the Protonate in pH box makes it possible to decide which pH value to use to protonate the chosen protein. The pH values can be from 0 to 14.
By clicking on the “Optimise structure” button, the optimisations start and the user is redirected to the Calculation progress page.
The Main page also offers three use cases of how the quality of structures is improved by FFFold optimisation.
This page informs a user about the optimisation job status. In addition to the optimisation identifiers (UniProt number and pH), the page shows the job status (queued or running).
When the optimisation is complete, the user is redirected to the Optimisation results page. The address for the Optimisation status page and Optimisation results page is identical so that the user can turn off the page and get back to the results. The optimisation status page and Optimisation results page are in the form fffold.biodata.ceitec.cz/results?ID={UniProt_number}_{pH} (e.g., https://fffold.biodata.ceitec.cz/results?ID=L8BU87_7.2)
First, the structure is downloaded from the AlphaFold DB in PDB format. Then the structure is protonated by PROPKA3 [Olsson2011] at the default pH or user-defined pH. PROPKA3 is run with default settings. Subsequently, protein regions predicted with confidence less than 90 are optimized by physics-based generic partially polarizable force field GFN-FF [Spicher2020] implemented in the xtb software [Bannwarth2021]. In protein regions predicted with higher confidence than 90, only the added hydrogens are optimized. Solvent is included during optimisation via the ALPB model [Ehlert2021]. The calculation is accelerated by a divide-and-conquer approach which results are comparable to the optimisation of whole protein structure with constrained α-carbons. During the optimisation, the structure is not optimised all at once but is divided into subparts, where each subpart is optimised separately [Schindler2022]. To provide the user with the optimised structure also in mmCIF format, the structure in mmCIF format is downloaded from the AlphaFold DB and the new atom coordinates are written into it after optimisation. The optimization time is usually a maximum of 10 minutes.
This page includes optimisation process identifiers, visualisation of original and optimised structures and a download data section.
This information consists of the UniProt structure code and pH. An example of the information is the following:
FFFold integrates Mol* viewer [Sehnal2021] to show original and optimised structure. The user has a choice of Cartoon, Surface and Ball & Stick visualisation modes. In the case of Cartoon and Surface, only the optimised structure is shown. In the case of Ball & Stick, the user can also visualize the non-optimised structure, which is colored gray. Visualization of the original and optimised structure in Ball & Stick mode is the default setting.
In addition to colouring according to the structure, the structure can also be coloured according to AlphaFold model confidence (according to pLDDT score).
For an efficient visualization of the individual residues and a quick check of the convergence of their optimisation (see Limitations) there is a list of one-letter shortcuts for each residue that can be clicked to zoom in on. Residues whose RMSD between optimised and non-optimised variants is greater than 1 are colored dark blue. Other converged residues are colored light blue and non-converged residues are colored orange.
The ”Download optimised structure” button allows the user to download a ZIP file containing four files, specifically:
- PDB file contains coordinates of the original structure from AlphaFold DB. The file is in the form {UniProt number}.pdb
- PDB file contains coordinates of the optimised structure by FFFold. The file is in the form {UniProt number}_optimised.pdb
- mmCIF file contains coordinates of the original structure form AlphaFold DB. The file is in the form {UniProt number}.cif
- mmCIF file contains coordinates of the optimised structure by FFFold. The file is in the form {UniProt number}_optimised.cif
Exceptionally, the optimisation may fail to convert. In case of non-convergence of the optimisation of residuum, the default xtb optimiser engine is replaced by the Approximate Normal Coordinate Rational Function optimiser and the substructure is optimised a second time. If the second optimisation does not converge either, the residuum is left. The user can find out whether a residue is optimised by the background color of the one-letter shortcut in the residue list in the Mol* visualizer interface.
| OS | Version | Chrome | Firefox | Edge | Safari |
|---|---|---|---|---|---|
| Linux | Ubuntu 23.10 | 120 | 121 | 120 | n/a |
| Windows | Windows 10 | 119 | 121 | 119 | n/a |
| MacOS | Sonoma 14.2 | 120 | n/a | n/a | 17.2 |
If you encounter an error or have an idea for an improvement, please send a report to Ondrej Schindler ([email protected]) or open a GitHub issue. Thank you!
- [Jumper2021] Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., ..., Hassabis, D. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596(7873), 583-589.
- [Varadi2022] Varadi, M., Anyango, S., Deshpande, M., Nair, S., Natassia, C., Yordanova, G., ..., Velankar, S. (2022). AlphaFold Protein Structure Database: massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic acids research, 50(D1), D439-D444.
- [UniProt2019] UniProt Consortium. (2019). UniProt: a worldwide hub of protein knowledge. Nucleic acids research, 47(D1), D506-D515.
- [Olsson2011] Olsson, M. H., Søndergaard, C. R., Rostkowski, M., Jensen, J. H. (2011). PROPKA3: consistent treatment of internal and surface residues in empirical pKa predictions. Journal of chemical theory and computation, 7(2), 525-537.
- [Sehnal2021] Sehnal, D., Bittrich, S., Deshpande, M., Svobodová, R., Berka, K., Bazgier, V., ... & Rose, A. S. (2021). Mol* Viewer: modern web app for 3D visualization and analysis of large biomolecular structures. Nucleic Acids Research, 49(W1), W431-W437.
- [Spicher2020] Spicher, S., & Grimme, S. (2020). Robust atomistic modeling of materials, organometallic, and biochemical systems. Angewandte Chemie International Edition, 59(36), 15665-15673.
- [Bannwarth2021] Bannwarth, C., Caldeweyher, E., Ehlert, S., Hansen, A., Pracht, P., Seibert, J., ... & Grimme, S. (2021). Extended tight‐binding quantum chemistry methods. Wiley Interdisciplinary Reviews: Computational Molecular Science, 11(2), e1493.
- [Ehlert2021] Ehlert, S., Stahn, M., Spicher, S., & Grimme, S. (2021). Robust and efficient implicit solvation model for fast semiempirical methods. Journal of Chemical Theory and Computation, 17(7), 4250-4261.
- [Schindler2023] Schindler, O., Berka, K., Cantara, A., Křenek, A., Tichý, D., Raček, T., & Svobodová, R. (2023). αCharges: partial atomic charges for AlphaFold structures in high quality. Nucleic Acids Research, gkad349.