AttnSeq-PPI Protein–Protein Interaction

Tutorial

A step-by-step guide to running predictions and reading the results.

AttnSeq-PPI predicts the probability that two proteins interact, directly from their amino-acid sequences, and — when you submit many proteins — assembles the results into an interactive interaction network. There are three ways to run it, all reached from the Predict PPI page. Pick a mode with its radio button, provide the sequences, and click Predict Interaction.

On this page

  1. Before you start — input rules
  2. Mode 1 — Pairwise interaction (one pair)
  3. Mode 2 — Between two groups (two files)
  4. Mode 3 — Within a single group (one file)
  5. Reading the results
  6. Input limits & troubleshooting

Before you start — input rules

Every mode accepts protein sequences only (the 20 standard amino-acid letters, plus the ambiguity codes B, Z, X, U, O). DNA/RNA, numbers, and other symbols are rejected. For file modes, use FASTA format — each entry is a > header line followed by the sequence.

New to the tool? Every mode has an Insert Example … button that fills the form with ready-made example data, and a Download Example … button if you'd rather inspect the file first. Use these to see a full run in one click.

Mode 1 — Predict Pairwise Protein–Protein Interaction

Use this when you have exactly one pair of proteins and want a single interaction probability.

  1. On the Predict page, select Predict Pairwise Protein-Protein Interaction (it is active by default).
  2. Paste the first sequence into Protein Sequence 1 and the second into Protein Sequence 2. Or click Insert Example Sequences to fill both automatically.
  3. Click Predict Interaction.
  4. The result page shows Probability of interaction as a percentage — higher means the pair is more likely to interact.
Each sequence may be pasted as a plain amino-acid string (headers are not required in this mode). Whitespace and line breaks are ignored automatically.

Mode 2 — Predict PPI Among Different Group

Use this to test every protein in one group against every protein in another — an all-vs-all comparison between two FASTA files (group 1 × group 2).

  1. Select Predict PPI Among Different Group to activate this section.
  2. Choose a FASTA file for Protein Sequences Group 1 and another for Protein Sequences Group 2. Or click Insert Example Groups to load both example files at once.
  3. Click Predict Interaction and wait while the model runs (a loading spinner is shown).
  4. You get an interaction table for every group-1 × group-2 pair plus an interactive network — see Reading the results.
The number of predictions is group 1 × group 2. To keep jobs fast this must stay within the pair limit — see Input limits.

Mode 3 — Predict PPI Among Single Group

Use this to test every protein in a single file against every other protein in the same file — all internal pairs of one group.

  1. Select Predict PPI Among Single Group.
  2. Choose your FASTA file under Select File Containing Protein Sequences, or click Insert Example Group to load the example file.
  3. Click Predict Interaction and wait for the run to finish.
  4. The results (table + network + downloads) appear exactly as in Mode 2, but the pairs are the unique combinations within your one file.
For a file of n proteins this runs n × (n − 1) / 2 predictions — every unique pair, without repeats.

Reading the results

Interaction table

Modes 2 and 3 produce a scrollable table with one row per predicted pair:

ColumnMeaning
Protein1_name / Protein2_nameThe FASTA header IDs of the two proteins in the pair.
id_1 / id_2Short node labels used in the network (e.g. A1, B3).
Interaction_Probability(%)Predicted probability that the two proteins interact.

Click Download TSV Table to save the full table as a tab-separated file you can open in Excel or a script.

Interactive network

Below the table, predicted interactions are drawn as a network you can drag, zoom, and rearrange. A link is drawn between two proteins only when their predicted probability is ≥ 50%, so the network highlights the confident interactions. In two-group mode the two groups are shown in different colours. Click Download Network Visualization to save the network as a standalone interactive HTML file.

Results are generated per run and are available to download for a limited time — save the TSV and network files if you want to keep them.

Input limits & troubleshooting

To keep the shared server responsive, each submission is checked against these limits. If one is exceeded you'll see a clear message explaining what to fix.

LimitValue
File type.fasta, .fa, or .txt
File sizeup to 2 MB per file
Sequences per fileup to 20
Sequence length20 – 1500 amino acids
Total pairs per jobup to 400 predictions
Allowed charactersamino-acid letters only (20 standard + B, Z, X, U, O)

Common messages and how to resolve them:

  • “…contains an invalid character” — the input isn't a protein sequence. Remove numbers, spaces, gaps, or DNA/RNA bases.
  • “No protein sequences found” — the file isn't valid FASTA; make sure each entry starts with a > header line.
  • “…has N sequences” / “would compute N pairwise predictions” — reduce the number of sequences so the job stays within the limits above.
  • “Too many requests” / “server is busy” — the tool runs one job at a time and is rate-limited; wait a moment and try again.