Predict a pair or batch
Open Interaction Prediction. Choose Single pair or Batch CSV. Paste sequences or upload a CSV, resolve input errors, review truncation warnings, and submit. Save the private link. Read probability alongside threshold and model version.
Retrieve candidates
Open Natural Sequence Retrieval. Select the query type, check the automatically selected opposite library, enter one sequence, choose Top K, and submit. Review similarity and source. Use Predict this pair to prepare a separate classification job.
Prepare input
Use plain sequence text or one FASTA record per sequence field. Batch CSV uses one pair per row and the documented column names. The server checks every row and rejects the whole batch if any row is invalid.
Long sequences
ESM processes at most 512 tokens including special tokens; AAindex uses the first 100 residues. The accepted length limit of 10,000 residues does not mean all residues enter the model.
Queue and return
CPU jobs run one at a time. Closing the page does not cancel an accepted job. Return using the complete private link before expiry. Keep the link private.
Shared submission limits
Each IP address may have up to two unfinished jobs and ten accepted submissions in a rolling hour. People sharing a network exit share these limits.
CSV exports
Potential spreadsheet formulas in text identifiers are prefixed with an apostrophe for safe CSV export. JSON preserves the original identifiers.
Fix errors
Correct the reported row and field for input errors. For a busy queue, wait and resubmit. For a failed job, read the error and submit again if appropriate. An expired result must be recomputed from your original input.
FASTA / CSV
Whitespace and alignment gaps are removed, letters are uppercased, and one terminal * is removed.
>format_example ACDEFGHIKLMNPQRSTVWY
pair_id,human_seq,viral_seq format_1,ACDEFGHIKLMNPQRSTVWY,MKTAYIAKQRQISFVKSHFS