NCBI BLAST compares a DNA or protein sequence with millions of known sequences in the NCBI database and finds the most similar ones. This workflow guides you through choosing the right type of BLAST, running a search or an alignment, and reading the results.

1.1Choosing BLASTStep 1

Go to NCBI BLAST. The type of BLAST depends on what your sequence is and what you want to compare it with:

Translated programs read DNA in all six reading frames, so they find proteins even when you do not know the gene structure.

1.2Running BLASTStep 2

BLAST can be run in two ways:

To align many sequences together (for example, for a phylogenetic tree), use a multiple alignment tool such as Clustal Omega or MAFFT instead: BLAST compares your query with each sequence separately.

1. Enter your query. Paste your sequence (FASTA format or bare sequence) or an NCBI accession number, or upload a file.

   

   

2. Choose the database (searches only):

   Database	Contains	Use it to
   nr (non-redundant)	Almost all protein sequences in NCBI	Run the broadest search
   refseq_protein	Curated reference proteins	Get cleaner, less redundant hits
   swissprot	Manually reviewed proteins (UniProtKB/Swiss-Prot)	Find well-characterised proteins
   pdb	Proteins with an experimental 3D structure	Find structures to compare with your AlphaFold model

   These are blastp databases; blastn uses nucleotide databases such as core_nt or refseq_rna.

3. Optional: limit the organism. Type a species or group in Organism (for example, Oryza sativa or Fungi) to restrict the search, or tick Exclude to remove it.

4. Click BLAST. Searches take from seconds to a few minutes. Bookmark the results page or note the Request ID (RID): results are kept on the NCBI server for about 36 hours.

1.3Reading the resultsStep 3

The results page has four tabs:

1.3.1Read the Descriptions table

1.3.2Read the scores together

1.3.3Check the alignment

Check the alignment of your top hits in the Alignments tab (shortened example):

Query  1  MKCTNIFLTIGLVFLALLLGSQ  22
          MKC+NIFL IG+VFLALLLG+Q
Sbjct  1  MKCSNIFLAIGIVFLALLLGAQ  22

Above each alignment, Identities, Positives and Gaps summarise these counts.

1.4Troubleshooting

• No significant hits: check that the program matches your sequence (DNA into blastn, protein into blastp), remove any organism limit, or try a broader database such as nr.
• Error about the sequence: remove numbers, spaces or non-sequence characters, and keep only one header line starting with >.
• Too many hits from the same species: use the Organism field to exclude it, or switch to refseq_protein or swissprot.
• Top hits are all “hypothetical” or “uncharacterized” proteins: search swissprot to find characterised proteins.
• The search is slow: shorter queries and smaller databases run faster; keep the page open or save the RID.

1.5References

• Camacho C. et al. (2009). BLAST+: architecture and applications. BMC Bioinformatics 10: 421.
• NCBI BLAST help and documentation
• AI in Genomics 2026, Practical 3: Compare.

Genome annotation predicts where the genes are in an assembly and assigns a function to each of them. This chapter describes three ways to annotate an assembly: Prokka on the command line (2.1), Prokka in your browser on the Galaxy EU server (2.2), and the RAST web server (2.3).

2.1Prokka CLI

Command-line interface

Prokka annotates a bacterial, archaeal or viral assembly in 2–20 minutes on a laptop. It takes an assembly (FASTA or FNA) as input and predicts CDS, rRNA, tRNA, tmRNA and ncRNA features, then assigns products by matching against curated protein databases.

2.1.1Prokka installation and setup

Conda is the route that works on both Linux and macOS and pulls in all 20+ dependencies (BLAST+, Prodigal, Aragorn, Barrnap, HMMER, tbl2asn). Install Miniconda first if you do not have it.

# add the channels once, in this order
conda config --add channels defaults
conda config --add channels bioconda
conda config --add channels conda-forge

# install into its own environment (recommended)
conda create -n prokka_env -c conda-forge -c bioconda prokka
conda activate prokka_env

Alternatives if you do not use conda:

# macOS, Homebrew
brew install brewsci/bio/prokka

# Ubuntu / Debian
sudo apt-get install prokka

# Docker, no local dependencies at all
docker pull staphb/prokka:latest

Then index the databases and confirm the install:

prokka --setupdb      # builds the BLAST indices, run once after install
prokka --version      # should print e.g. prokka 1.14.6
prokka --listdb       # shows which kingdom and genus databases are available

If prokka --version fails with a Perl module error, the usual cause is a missing dependency outside conda; reinstalling inside a clean conda environment fixes it in almost every case. Remember to run conda activate prokka_env in every new terminal session.

2.1.2Annotation

Prokka takes the assembly as a positional argument: there is no input (-i) flag.

1. Open a terminal.

   • On Ubuntu, Ctrl+Alt+T.
   • On macOS, “Terminal” from “Applications › Utilities”.

2. Check where you are. This prints the current working directory, so you know what any relative path will be measured against.

   pwd

3. Copy the path of the folder holding your assembly. In the file manager, right-click the folder and choose “Copy path”.

   • Ubuntu: Ctrl+L in Files shows the path.
   • macOS Finder: right-click and hold Option, then “Copy as Pathname”.

4. Move into that folder. Type cd, a space, then paste. You will see something like this:

   cd /home/user/projects/genome_assembly

   OR drag and drop the folder onto the terminal window, which also pastes its path. Quote the path if it contains spaces.

5. List the files and confirm the assembly is there using ls, which lists all the files in your directory.

   ls

6. Run Prokka.

   prokka contigs.fasta --outdir results_prokka --prefix my_genome --compliant

What each part does:

• contigs.fasta: your assembly file.
• --outdir results_prokka: the folder to create for results. It must NOT already exist, or Prokka stops. If it does, add --force to your command to overwrite.
• --prefix my_genome: names every output file my_genome.gff, my_genome.faa and so on. Without it they are named after today’s date, which gets confusing fast.
• --compliant: enforces GenBank/ENA/DDBJ submission rules. It is shorthand for --addgenes --mincontiglen 200 --centre XXX, so it drops contigs under 200 bp and adds gene features alongside each CDS.
• Optional: --cpus 4: threads to use; --cpus 0 uses every available core.

A typical bacterial genome finishes in 2–20 minutes. Progress is printed to the terminal and mirrored in the .log file.

2.1.3Optional arguments worth adding

Prokka accepts single or double dashes interchangeably (-compliant and --compliant both work). prokka --help lists everything.

A fuller command for a known organism:

prokka contigs.fasta --outdir results_prokka --prefix ecoli_K12 \
       --genus Escherichia --species coli --strain K12 --usegenus \
       --locustag ECO --centre UoT --compliant \
       --cpus 8 --rfam

2.1.4Annotation outputs

Every file shares the --prefix you set, so with --prefix my_genome you get my_genome.gff, my_genome.faa and so on.

Quick checks after a run

cd results_prokka
cat my_genome.txt                  # feature counts
head -3 my_genome.tsv              # table header and first rows
grep -c ">" my_genome.faa          # number of proteins predicted

As a rough guide, a typical bacterial genome yields around 900 CDS per Mb. Far fewer usually means a fragmented assembly or the wrong --kingdom.

2.1.5Resources and references

• Seemann, T. (2014). Prokka: Rapid prokaryotic genome annotation. Bioinformatics, 30(14), 2068–2069. https://doi.org/10.1093/bioinformatics/btu153
• Prokka on GitHub: official documentation, full option list and output specification
• Prokka on Bioconda: installation recipe and current version

2.2Prokka GUI

On the Galaxy EU server

This workflow guides you through the annotation of your genome using the web-based version of Prokka, which is available on the Galaxy EU Server.

2.2.1Overview of Galaxy EU

The European Galaxy Server (usegalaxy.eu) is a public web platform for accessible, reproducible, and transparent computational research, primarily used for data analysis in fields like bioinformatics, genomics, and climate science.

Users can interact with thousands of specialized scientific tools via a graphical web interface rather than writing code.

2.2.2Running Prokka on the Galaxy EU Server

Set up your Galaxy EU account

1. Open usegalaxy.eu in any browser.
2. Click Login or Register in the top menu and create a free account. You can run tools anonymously, but an account keeps your histories, gives a larger storage quota and lets Galaxy email you when a job finishes.
3. Start a new history for this analysis: in the “> History” panel on the right, click the + icon and rename it (click the name) to something like Prokka – isolate 12. A history is Galaxy’s project folder; every file you upload and every result appears there as a numbered dataset.

The screen has three panels: tools on the left, the working area in the centre where tool forms open, and your history on the right.

Search for Prokka

1. In the search tools box at the top of the left panel, type Prokka.
2. Click Prokka – Prokaryotic genome annotation. The tool form opens in the centre panel.
3. Check the version button at the top of the form and use the newest one listed (1.14.6 builds at the time of writing), so results match the command-line version.

You can open the form now, but it only becomes useful once your genome is uploaded: the input dropdown lists datasets from your current history.

2.2.3Upload your genome

This is where most participants get stuck. The upload itself is easy; the trap is that Galaxy must recognise the file as fasta, or it will not appear in Prokka’s input list.

Upload the file

1. Under Contigs to annotate, click on “...”. Click the Upload icon (an arrow pointing up) at the bottom left of the panel. The upload dialog opens.
2. Add your assembly one of two ways:
   • Choose local file: pick the contigs FASTA from your computer.
   • Paste/Fetch data: paste a URL, for example an NCBI or Zenodo link, and Galaxy downloads it directly.
3. Set the Type column to fasta. It defaults to auto-detect, which usually works but sometimes labels an assembly as txt. Setting it yourself removes the guesswork. Leave Genome as ?.
4. Click Start, then Close the dialog. You do not need to keep it open.

Galaxy decompresses .gz and .zip files automatically, so a gzipped assembly is fine as it is. Upload the assembly, not the raw reads (.fastq): Prokka annotates contigs.

Check it arrived correctly and run the tool

The dataset appears at the top of your history and changes colour as it goes: grey queued, orange running, green done, red failed.

1. When it is green, this means that the upload is done, NOT the annotation. Click its name to expand it and check the number of your upload (your first upload would be 1).
2. Click the selection box under Contigs to annotate and choose the number that corresponds to your upload. In case you upload multiple genomes at once, this tells the tool which genome to annotate.
3. Click Run Tool at the top right of the central panel.
4. Output files will start showing in your history: grey (queued), orange (running), green (complete), or red (failed).
5. Once complete, you may download the output file of your choice using the save icon that is found in each green box. OR click the eye icon to preview the contents; you should see > header lines followed by sequence.

2.2.4Troubleshooting

2.2.5Sources

• Galaxy EU: the server, tool search and upload
• Galaxy Training: genome annotation with Prokka: the official step-by-step tutorial, useful as a companion exercise

2.3RAST

Rapid Annotation using Subsystems Technology

The Rapid Annotation using Subsystems Technology (RAST; rast.nmpdr.org) annotates an uploaded assembly on a remote server and returns it organised into subsystems (curated functional groups of genes), which you browse in the SEED Viewer. There is nothing to install, but jobs are queued and typically take a few hours to a day.

2.3.1Create a RAST account

Accounts are approved by a human, so register a few days before you need results.

1. Go to rast.nmpdr.org.
2. Click Register for a RAST account in the left-hand menu.
3. Fill in the form: name, institutional email, institution, and a short note on what you intend to annotate. An academic or institutional email address is approved faster than a personal one.
4. Submit, then wait for the approval email. This usually arrives within a working day, occasionally longer.
5. Log in from the RAST home page with the username and password you chose.

One account covers unlimited jobs, and your genomes stay private to you unless you explicitly make them public.

2.3.2Upload a new job

1. Log in and click Upload New Job in the left menu.
2. Choose the sequence file. Click Browse and select your assembly: a contigs FASTA (.fasta, .fna, .fa), optionally gzipped. Upload the assembly, not raw reads; RAST does not assemble.
3. Click Use this data to move to the settings page.
4. Fill in the organism taxonomic details.
   • Taxonomy ID: enter the NCBI taxid if you know it and click the lookup button; it auto-fills the rest. Otherwise leave it and type the fields manually.
   • Domain: Bacteria, Archaea or Virus.
   • Genus, species, strain: as precise as you can be. If unidentified, a genus plus “sp.” and your isolate code works.
   • Genetic code: 11 for most bacteria and archaea, 4 for Mycoplasma, Spiroplasma and relatives.
5. Choose the annotation scheme. Select RASTtk, the current toolkit and the default. Classic RAST is kept only for reproducing old jobs.
6. Set the options. The defaults are sensible; the ones worth a thought:
   • Automatically fix errors: leave ticked.
   • Fix frameshifts: tick it for draft assemblies, where sequencing errors create false frameshifts.
   • Build metabolic model: tick it if you want a draft flux-balance model in ModelSEED.
   • Backfill gaps: a second pass to catch genes missed on the first; harmless to leave on.
7. Click Finish the upload. You get a job ID immediately, and an email when the job completes.

Runtime depends on the queue and on genome size. Check progress under Jobs Overview, which shows each stage with a coloured status per stage.

2.3.3Find and read the results

When the email arrives, go to Jobs Overview and click the job, or use My Jobs. Two buttons matter: View details, which opens the SEED Viewer, and Download, which exports the annotation as GenBank, GFF3, EMBL, amino-acid FASTA, nucleotide FASTA or an Excel spreadsheet.

Clicking View details lands you on the organism genome metrics: contig count, genome size, GC content, number of coding sequences, number of RNAs, and the proportion of genes placed in subsystems.

To browse the actual annotation, click Browse annotated genome in SEED Viewer.

Subsystem statistics: the two views of the same data

A subsystem is a curated set of genes that together carry out one biological process: a pathway, a complex, a transport system. The subsystem statistics panel shows them two ways, side by side.

The two are linked: clicking a pie slice filters the table, and the table is where you drill down. Use the pie to orient yourself, the table to work.

Subsystem coverage is typically 40–60% of genes. The rest are real genes that are simply not in a curated subsystem yet, most of them hypothetical proteins. A low percentage is not a failed annotation.

2.3.4Choose a subsystem in the table

The table is a three-level hierarchy. You expand your way down to a named subsystem, and that subsystem gives you its gene list.

1. Pick a category. The top-level rows are broad: Amino Acids and Derivatives, Carbohydrates, Cell Wall and Capsule, Virulence · Disease · Defense, Membrane Transport, and so on. Click the category name, or the matching slice in the circular diagram.
2. Expand to a subcategory. Inside Carbohydrates, for example, you get Central carbohydrate metabolism, Fermentation, Monosaccharides, and others.
3. Click the subsystem itself. Inside Central carbohydrate metabolism you might pick Glycolysis and Gluconeogenesis or TCA Cycle. The subsystem name is the clickable link.
4. Read the subsystem page. It has two parts:
   • The functional roles table: each role in the pathway, and which of your genes fills it. Blank rows are informative: a missing role may mean an incomplete pathway, or a gene the annotation missed.
   • The gene list, with a feature ID for each, of the form fig|83333.1.peg.1234. peg stands for protein-encoding gene; the number is the gene’s index in your genome.
5. Click a feature ID to open that single gene.

A practical shortcut for a workshop exercise: Virulence · Disease · Defense › Resistance to antibiotics and toxic compounds gives participants an immediately interesting gene list in any isolate.

2.3.5One gene: neighbourhood and sequences

Clicking a feature ID (fig|83333.1.peg.1234) opens the Annotation Overview page for that gene.

The genetic environment

Scroll to the Compare Regions panel. It draws your gene as a red arrow in the centre of its contig, with the genes flanking it on either side, and stacks the equivalent region from closely related genomes underneath.

How to read it:

• Arrow direction is the strand, so you can see at a glance which neighbours are co-oriented and might sit in one operon.
• Colour equals homology. Genes sharing a colour across rows are orthologues; your red gene’s colour group tracks it through every genome shown.
• A conserved block of the same colours in the same order across many rows is a strong hint of a functional unit: an operon or a gene cluster.
• A break in synteny (neighbours that differ from the related genomes) often marks a mobile element, a genomic island, or a horizontal transfer event.
• Hover over any arrow for its function; click it to jump to that gene’s own page. The controls above let you widen the window and change how many genomes are compared.

The FASTA sequences

On the same page, the sequence section gives both forms of the gene:

Both display as plain FASTA with the feature ID in the header, so you can copy-paste straight into BLAST. There are also links out to run BLAST directly, and to the gene’s entry in related databases.

For many genes at once, do not scrape this page: go back to the job page and use Download to export the whole annotation as amino-acid or nucleotide FASTA.

2.3.6Sources

• RAST server: registration, job submission and the SEED Viewer
• BV-BRC: the successor platform, running the same RASTtk annotation service

GetGenome supports the sequencing of your organism of interest, providing an assembly file (.fasta) and a protein sequence file (.faa) that is generated from the annotation of the genome. This workflow guides you through the steps to model your protein of interest (POI) using the .faa file and AlphaFold3 (AF3) Server.

3.1Overview

AlphaFold 3 Server is an advanced artificial intelligence model developed by Google DeepMind and Isomorphic Labs that predicts the 3D structures and interactions of life’s cellular molecules (proteins, DNA, RNA, chemical modifications, and small molecule ligands or ions).

3.2The input file

The protein sequence file (.faa) contains the protein sequences from your genome, with identifiers and, where possible, predicted functions. You do not need to identify the proteins yourself; you only need to find your protein of interest in the file. (A .faa file is one of the outputs of genome annotation; see 2. Genome annotation.)

3.2.1Open your protein sequence fileStep 1

Open the .faa file in a text editor (for example “Notepad”, “Sublime Text”). Do not use Microsoft Word or Microsoft Excel as they do not support such file formats.

3.2.2Locate your amino acid sequenceStep 2

AlphaFold 3 Server requires the primary amino acid sequence of your protein as the input. The protein file links each amino acid sequence to an identifier.

1. Use Find (Ctrl+F or Cmd+F) and type your protein’s name or symbol. A matching line looks like this (shortened):

   contig_12 annotation gene 803242 803583 . - . ID=g04521;Name=AVR-Pik;product=AVR-Pik effector
   contig_12 annotation mRNA 803242 803583 . - . ID=g04521.t1;Parent=g04521

   Note the identifier of the mRNA line (here g04521.t1): the protein sequence is usually stored under this identifier.

2. If the gene has several mRNA lines (for example g04521.t1 and g04521.t2), the gene has more than one predicted version of its protein. Use the primary one: the version marked as canonical or representative in the annotation if it is marked, otherwise the first (.t1), which is usually the longest.

3. Copy the sequence. Copy the primary amino acid sequence, which will end before the next >:

   >g04521.t1
   MKCTNIFLTIGLVFLALLLGSQAEAETFLTIGLVFLALL
   >g03881.t1

Optional (recommended): Save the amino acid sequence as a new text file, for example my_protein.faa.

3.2.3Checkpoints

Before modelling, look at the sequence:

3.2.4Troubleshooting

• Search finds nothing: gene names in the annotation may differ from the ones you know. Try the predicted function, or use the BLAST route:

If your protein is not named in the file, take its amino acid sequence and search it with blastp (see 1. Sequence alignment). The top hit gives you the identifier.

• Several different genes match: your protein may belong to a family with several members in the genome. Use BLAST against a known version to pick the closest one.
• Identifier not found in the proteins file: search for the gene identifier instead of the mRNA identifier (for example g04521 rather than g04521.t1); naming conventions vary.

3.3Input to AlphaFold 3 ServerStep 3

1. Sign in at alphafoldserver.com with your Google account.
2. In the first entity box, choose Protein as the molecule type.
3. Paste your sequence without the header line (the line starting with >). Remove any trailing * or ..
4. To model your protein together with other molecules, click Add entity and add each one: another protein, DNA, RNA, a ligand or an ion.
5. Give the job a descriptive name.
6. Leave the seed* on automatic for a first run.
7. Preview the job, check that every entity and copy number is correct, then Confirm and submit job.

The server limits the total size of each job and the number of jobs per day. Both are shown on the server.

* A seed is the random starting number for a prediction: the same seed reproduces the same result, and different seeds give alternative predictions whose agreement tells you how reliable the structure is.

3.3.1Number of copies

Each entity has a Copies field. It sets how many copies of that same molecule are in the model.

Set more than one copy only when there is evidence that the protein works that way, from the literature or from known structures of related proteins. AlphaFold will build a complex from whatever copies you give it, and the result can look plausible even when that complex does not exist in nature.

3.4Understand the outputStep 4

When the job finishes, open it from the job list. The results page shows:

• the 3D model, coloured by confidence (see below);
• the pTM and, for complexes, ipTM scores;
• the green plot (see below);
• a selector to switch between the five models AlphaFold produces for each job.

Click Download to get a zip file:

The .cif files open in UCSF ChimeraX (see 4. Protein visualization).

3.5Interpreting the confidence scores

Read the confidence scores before looking at any structural detail. A model without its confidence scores should not be interpreted.

3.5.1Colours on the model (pLDDT)

The model is coloured residue by residue by how confident AlphaFold is in the local structure:

Orange regions are often flexible or disordered parts of the protein. AlphaFold still draws them, sometimes as long loops or isolated helices, but their shape carries no information.

3.5.2pTM

pTM is a single score from 0 to 1 for the overall fold of the whole model.

3.5.3ipTM

ipTM appears only when the job has more than one entity or copy. It scores, from 0 to 1, how confident AlphaFold is in the way the entities are placed relative to each other.

pTM and ipTM are less reliable for very small proteins and short chains; for these, rely more on the colours and the green plot.

3.5.4The PAE plot (Predicted Aligned Error)

The PAE plot shows how confident AlphaFold is in the relative position of every part of the model against every other part.

• Both axes list the residues of the model, in order, chain after chain. When there are several chains, lines on the plot mark where one chain ends and the next begins.
• Each point shows the expected error in the position of the residue on one axis when the model is aligned on the residue on the other axis.
• Dark green means low expected error (confident); light or white means high expected error (not confident).

How to read common patterns:

The colours on the model and the green plot answer different questions. A region can be dark blue on the model, meaning its local structure is confident, while the green plot shows its position relative to another domain or chain is uncertain.

3.5.5Reading the scores together

A confident model is a well-founded hypothesis about your protein’s structure, not proof. Use it to decide which experiments are worth doing.

3.6Troubleshooting

• “Invalid sequence”: a header line, a * or ., spaces, or letters that are not amino acids were pasted.
• Job too large: model a single domain, or reduce the number of copies or partners.
• Daily limit reached: the counter resets; submit the job the next day.
• Everything is orange: check the sequence (Step 2 checkpoints) before re-running.

3.7References

• Abramson J. et al. (2024). Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature.
• Maqbool A. et al. (2015). Structural basis of pathogen recognition by an integrated HMA domain in a plant NLR immune receptor. PMID 26304198.
• Posbeyikian A., Toghani A., McClelland, Kamoun S., Sugihara Y., Contreras (2026). Fold on Tight: Ten Things to Know About AI Structure Prediction in Plant Immunity.

UCSF ChimeraX is a free program to visualise and analyse 3D protein structures. This workflow guides you through installing ChimeraX and learning its basic commands.

4.1InstallationStep 1

Download ChimeraX from the official download page, choosing the version for your operating system. It is free for academic and non-commercial use: accept the licence to start the download. Install the latest version, as AlphaFold 3 files need a recent one.

Open ChimeraX and type version in the command line at the bottom of the window: the version number appears in the Log.

4.1.1Troubleshooting

• Installation is blocked: institutional laptops often need administrator rights; ask your IT department.
• macOS says the app cannot be opened: go to System Settings → Privacy & Security and click Open Anyway.

4.2Using ChimeraXStep 2

ChimeraX is used through menus and through commands typed in the command line. Every command and its result appear in the Log.

4.2.1Open a structure

Open a structure from the Protein Data Bank (PDB) with its four-character ID:

open 7B1I

Click [more info...] in the Log to see its metadata, such as the paper title and the experimental method.

4.2.2Point to parts of a structure

Use these symbols, combined from left to right (for example #1/b:1-10):

4.2.3Select, colour and display

Close the structure before the next step:

close #1

4.3Open your AlphaFold 3 modelStep 3

Unzip your AlphaFold 3 download (see 3. Protein modelling) and drag …_model_0.cif (the top-ranked model) into the ChimeraX window, or type its path:

open ~/Downloads/fold_my_job/fold_my_job_model_0.cif

4.3.1Colour by pLDDT

Blue = confident, orange = low confidence:

color bfactor #1 palette alphafold

4.3.2Load the PAE plot

Go to Tools → Structure Prediction → AlphaFold Error Plot and select your model. ChimeraX usually finds …_full_data_0.json; if not, click Browse, select it, then Open. The same, by command:

alphafold pae #1 file ~/Downloads/fold_my_job/fold_my_job_full_data_0.json

The PAE plot is interactive: drag a box over a region of the plot and it is highlighted on the 3D model. The plot window also recolours the model by pLDDT or by PAE domains.

4.4Analyse the interfaceStep 4

4.4.1Show contacts

Show contacts between chain A and chain B closer than 3.5 Å. Each contact is drawn as a line (pseudobond) coloured by PAE, from blue (confident) to red (not confident). The PAE plot must be loaded first (Step 3).

alphafold contacts #1/a to #1/b distance 3.5

To look at one residue only, here residue 20 of chain B within 4.5 Å of chain A:

alphafold contacts #1/b:20 to #1/a distance 4.5

4.4.2Export the contacts

Export the contacts to a text file, to sort them in Excel:

alphafold contacts #1/a to #1/b distance 3.5 outputFile AF3-contacts.txt

Each row is a pair of residues in contact and its PAE, for example /A:81 /B:54 17.50.

Residues in confident contacts are candidates to mutate when testing the interaction in the lab.

4.5Save your workStep 5

ChimeraX does not save automatically.

To record a 360° spin movie, run these four commands in order:

movie record
turn y 2 180
wait 180
movie encode ~/Desktop/spin.mp4

Pictures and movies can also be taken from Home → Images. Type pwd to see the folder where files are saved. On Windows, you can also write full paths, such as C:\Users\yourname\Desktop\image.png.

4.6Troubleshooting

• Command not recognised: check the spelling, or type help followed by the command name.
• Wrong residues coloured: check the model number (#), chain ID (/) and residue numbers in the Models panel and the Sequence viewer.
• The PAE plot does not load: select the full_data file with the same number as the model (model_0 with full_data_0).
• The contacts command gives an error: load the PAE plot first (Step 3).
• File not found: the path is wrong; drag the file into ChimeraX instead.

4.7References

• Meng E.C. et al. (2023). UCSF ChimeraX: Tools for structure building and analysis. Protein Science 32: e4792.
• ChimeraX User Guide
• AI in Genomics 2026, Practical 2: Interpret (slides).

Foldseek compares a protein structure with millions of known and predicted structures and finds the most similar ones, even when their sequences are unrelated. This workflow guides you through preparing your structure, running a search, reading the results, and aligning several structures with FoldMason.

5.1Before you startStep 1

Get everything you can from the sequence first: run BLAST (see 1. Sequence alignment) and InterProScan. Use Foldseek when the sequence gives few or no answers.

Foldseek needs a structure file in PDB or CIF format, either an experimental structure or a prediction (for example, …_model_0.cif from AlphaFold 3; see 3. Protein modelling). Your results are only as good as the structure you put in, so check it first:

To keep one chain and save it as a PDB file in ChimeraX (here, removing chain B):

open model_0.cif
delete /B
save my_protein.pdb

5.2Searching with FoldseekStep 2

Go to Foldseek Search.

1. Upload your structure with UPLOAD PDB, or use LOAD ACCESSION to load a structure by its identifier.

2. Choose the databases. Keep all of them for the broadest search, or untick the ones you do not need:

   Database	Contains
   PDB100	Experimental structures from the Protein Data Bank
   AlphaFold/Swiss-Prot	Predicted structures of manually reviewed proteins
   AlphaFold/UniProt50	Predicted structures covering most known proteins
   AlphaFold/Proteome	Predicted structures from selected model organisms
   CATH50	Classified protein domains
   Other databases	Predicted structures from metagenomes, viruses and other sources

3. Choose the mode:

   Mode	Compares	Use it to
   3Di/AA	Local similarity, using structure and sequence together	Run a fast, sensitive search for related proteins (default)
   TM-align	Overall (global) structural similarity	Test whether hits share the whole fold
   LoL-align	Local structural similarity	Find shared regions that suggest a common origin

4. Optional settings. Use Taxonomic filter to restrict hits to a group of organisms. Tick Iterative search to find more distant relatives.
5. Click SEARCH. Results usually appear within a few minutes. Bookmark the page, or download the results to reload them later with UPLOAD PREVIOUS RESULTS.

5.3Reading the Foldseek resultsStep 3

Results are listed per database, best hits first. Click a hit to see its alignment with your query and the two structures superposed.

5.3.1Read the scores together

A similar structure suggests, but does not prove, a shared origin or function. No single tool gives the whole answer: combine Foldseek with BLAST, domain annotation and the literature.

5.4Aligning many structures with FoldMasonStep 4

To compare several structures at once, for example your query and its best hits, use FoldMason, on the same website.

1. Upload at least two structures (PDB or mmCIF): click CLICK TO SELECT FILES or drag and drop them.
2. Click ALIGN.

The result is a structure-based multiple alignment: residues that occupy the same position in 3D are aligned, even when the sequences differ. Look for columns that are conserved across all structures, such as cysteines or glycines, which often hold the fold together.

5.5Troubleshooting

• The upload is rejected: check the file is PDB or mmCIF; re-save it from ChimeraX.
• Hits match a partner protein instead of yours: the file contains several chains; keep only your protein (Step 1).
• Many weak hits: switch to TM-align mode and focus on hits with TM-score above 0.5.
• No hits: try 3Di/AA mode with iterative search, and keep all databases selected.
• Hits only match a disordered or low-confidence region: remove low-pLDDT regions and search again.

5.6References

• van Kempen M. et al. (2024). Fast and accurate protein structure search with Foldseek. Nature Biotechnology 42: 243–246.
• Xu J. and Zhang Y. (2010). How significant is a protein structure similarity with TM-score = 0.5? Bioinformatics 26: 889–895.
• Foldseek on GitHub
• AI in Genomics 2026, Practical 3: Compare (slides).

Key terms used across the AI in Genomics workflows, grouped by topic and listed from A to Z within each topic.

No terms match your search. Try another word or choose All topics.