How to predict structures with AlphaFold

From Proteopedia

Jump to: navigation, search

In 2020, the AlphaFold project of Google's DeepMind team demonstrated a major breakthrough in predicting protein structure from sequence. Their success in the blind CASP competition astonished many experts. For an overview, see Theoretical models, bearing in mind "The Joys and Perils of AlphaFold"[1]. AlphaFold2 continued to have the highest success rate in the 2022 CASP 15 competition.

In July, 2021, DeepMind released AlphaFold as open source code. Subsequently, several Colabs became available offering free structure prediction for user-submitted protein sequences. These Google Colabs (collaboratories)[2]. enable users to submit sequences via web browser, executing the code in the Google cloud, using space private to each user, returning predicted structures.

Below are instructions for beginners who wish to predict structures. We recommend the "advanced" Colab by Sergey Ovchinnikov, Milot Mirdita and Martin Steinegger. Some of the options recommended below were gleaned from the 1 hour 46 minute video of presentations by Sergey Ovchinnikov and Martin Steinegger (August, 2021) for the Boston Protein Design and Modeling Club hosted by Chris Bahl.

Contents

First Check AlphaFold Database

Structure predictions for over 300,000 proteins are already available in the AlphaFold Database. If your protein is there, you don't need to proceed with the instructions below. Simply download the prediction from the Database.

Also check the AlphaFill Database, which has added ligands to appropriate AlphaFold predictions. Ligand positioning is approximate. See CAUTION provided by the AlphaFill team.

Single chain limitation

Initially, AlphaFold and ColabFold performed best with single chains[3], which may include one or a few domains. The instructions below were written before ColabFold was adapted to prediction of complexes or alternate conformations. If you are interested in complexes or alternate conformations, please see ColabFold instructions in the 2023 paper by Kim et al. [4]

Submitting A Sequence

First, check the AlphaFold Database for the protein of interest. If its structure has already been predicted there, download it, and skip to Interpreting Results below. Otherwise ...

Don't worry about any of the options not specifically mentioned below. Leave them at their default settings.
1. Obtain the sequence of the protein of interest, e.g. at UniProt. Click on the FASTA button above the sequence in UniProt. Copy only the sequence, excluding the FASTA header line that begins with ">".

2. Login with a google account at AlphaFold2_advanced. You can register for a free gmail account to use for login.

3. Paste in your sequence, making sure to completely replace the default sequence:

This input slot can accept sequences >1,000 amino acids, even though it is only one line. Sequence lengths of ~1,000 amino acids, or longer, may cause the Colab to fail, but can be predicted by submitting in two halves.[5] See also [5] and Joining AlphaFold predictions for halves of a molecule.

4. Enter a jobname in the slot below the sequence slot. The results.zip filename will begin with this jobname (but none of its contents include the jobname).

5. Scroll down to the section titled run alphafold, subsection Sampling options:

  • num_models, the number of models to be predicted, is 5 by default. You could reduce this to 3 if you are in a hurry.
  • max_recycles: Set this to 48 (or at least 12). The actual number of "recycles" performed will stop when the model has converged to the specified tolerance. The default of 3 recycles is often not enough for an optimal result.
  • tol (tolerance): Set this to 0.5 Å (or 1.0 to get a faster result). When a prediction differs from the previous "recycle" prediction by less than this value (RMSD in Å between alpha carbons), the recycles will stop.
  • num_samples (random seeds): Leave this at 1. Beware that if you increase this above 1, you will generate a number of models equal to the product of this value times num_models. This will proportionally increase the time to complete a result.


6. Open the Runtime menu at the very top of the page, and select Run all.
Image:AF2Adv-runall.png
Don't worry about the "Warning". It is just Google's disclaimer that they did not write the code you are about to execute. Click Run anyway.

Downloading Results

Do NOT close your AlphaFold2_advanced browser tab until the job is completed. It appears that you will lose your job if you close the browser tab. You will be warned if you inadvertently try.

When the job is completed, a dialog to download a zip file will appear automatically. (Sometimes you will be asked for permission to enable download first.)

Interpreting Results

Static images of backbone renderings of predicted models will appear in your web browser at the bottom of the section run alphafold as each is completed.

Estimated Reliability

Each predicted model has an average estimated reliability (pLDDT, predicted local distance difference test). >90 is likely accurate; <70 is low confidence. For more about interpreting these values, please see the AlphaFold Database FAQ.

Each residue has an estimated reliability of its position (0-100) in the PDB temperature column. BEWARE that high values mean high confidence, and low values mean low confidence. This is the INVERSE of crystallographic temperature values, where low values are good and high values are bad. Uploading your PDB file to FirstGlance in Jmol will automatically color each residue by its estimated reliability.

Visualization

FirstGlance in Jmol automatically colors its initial view of uploaded AlphaFold models by estimated reliability per residue (blue for high confidence, red for low confidence). After you go to other views or tools, you can always get back to this color scheme by clicking Reliability Estimates in the Views tab.

After uploading your predicted model to FirstGlance.Jmol.Org, you can easily visualize

  • Estimated reliability per residue
  • Secondary structure (Views tab)
  • Distribution of hydrophobic vs. polar residues (Views tab: integral membrane proteins will have large hydrophobic surfaces while soluble proteins will have hydrophobic cores revealed by the Slab button)
  • Distribution of charges (Views tab: nucleic acid binding sites will have clusters of positive charges)
  • Disulfide bonds (Tools tab)
  • Domain structure and positions of the ends of the polypeptide chain (Views tab: N -> C Rainbow)
  • Locations of functional sites by evolutionary conservation (see instructions at How_to_see_conserved_regions)

Intrinsic Disorder

Some models have high confidence in a folded domain, and low confidence in a segment that is not part of a compact domain. Low-confidence segments may be intrinsically disordered. It is useful to compare predictions of disorder with AlphaFold reliability estimates.

Relative Positions of Domains

If the predicted model has more than one domain, each domain may have high confidence, yet the relative positions of the domains may not. The estimated reliability of relative domain positions is in graphs of predicted aligned error (PAE) which are included in the downloadable zip file of results. For an explanation, see How should I interpret the relative positions of domains? in the AlphaFold Database FAQ.

Recycles For Convergence

You may be interested to note the number of recycles required for each model to converge to the specified tolerance. These numbers are not captured in the downloaded zip file.

The models will be ranked with number one having the highest estimated reliability (pLDDT). This is usually not in the order in which they were calculated. You might want to copy the ranking list, perhaps adding the number of recycles and final tolerance values:

model rank based on pLDDT              Recycles   Tolerance
rank_1_model_2_ptm_seed_0 pLDDT:62.46    10          0.33
rank_2_model_3_ptm_seed_0 pLDDT:59.59     9          0.47
rank_3_model_1_ptm_seed_0 pLDDT:55.63    12          0.52

Notice that the model predicted 2nd had the best estimated reliability (pLDDT), and that the model ranked 3rd did not quite achieve the specified tolerance of 0.5 Å RMSD after 12 recycles. (12 was specified as the maximum in this job.)

Also notice that, in this case, all 3 models have low confidence (pLDDT < 70), and are of questionable value.

Paying for Colab Pro

When I submitted a new job, after completing several large jobs (>500 residues), I was informed that a GPU could not be assigned. My access was temporarily restricted because of the free resources I had recently consumed.

In October, 2021, a subscription to Colab Pro was US $10/month. After I subscribed, all jobs I have submitted have been processed without restriction.

See Also

References and Notes

  1. Perrakis A, Sixma TK. AI revolutions in biology: The joys and perils of AlphaFold. EMBO Rep. 2021 Oct 20:e54046. doi: 10.15252/embr.202154046. PMID:34668287 doi:http://dx.doi.org/10.15252/embr.202154046
  2. Collaboratory FAQ at Google.
  3. Protein complex prediction with AlphaFold-Multimer, 2021, Evans et al. (DeepMind Team).
  4. Easy and accurate protein structure prediction using ColabFold, 2023, Kim et al. (DeepMind Team).
  5. 5.0 5.1 I had one sequence of length ~1,300. After it failed, I submitted it as two halves with a substantial overlap (~350 residues). The middle overlap of ~200 residues of the predicted structures superposed very closely with DeepView. I trimmed off the ends that superposed poorly, and superposed the two halves via the mid-overlap. By inspection, I chose pair of alpha carbons near the middle where the alpha carbon positions were nearly identical. I trimmed each half to this position, and "ligated" the two halves by combining the superposed half PDB files with a text editor. For further details, contact User:Eric_Martz.

Proteopedia Page Contributors and Editors (what is this?)

Eric Martz

Personal tools