What can I use DeepHVI for?
The two tasks support research on human-virus protein interactions. Predictions require experimental validation. Please cite the DeepHVI paper when using the results.
Why is my task queued?
A single CPU process handles tasks in submission order. Waiting time depends on the queue and the sequence encoding required.
What are the input limits?
Pairwise prediction accepts up to 100 pairs per request. Partner prediction accepts one sequence and returns 5, 10, or 20 candidates. Each sequence may contain up to 10,000 residues, and each request must not exceed 5 MiB.
Are long sequences processed in full?
No. ESM uses at most 1,024 tokens in total, including BOS/EOS (normally up to 1,022 residues), and AAindex uses the first 100 residues.
Is similarity an interaction probability?
No. Cosine similarity measures the match between a predicted partner embedding and a natural protein embedding. The two prediction tasks can be used independently.
Does partner prediction return newly generated sequences?
No. The model predicts a partner embedding and retrieves natural sequences. All returned sequences come from the specified reference-library version.
Do I need an account?
No. Access is through a private job link. Anyone with the complete link and its token can access the task.
How long are results retained?
Inputs and results are retained for seven days from submission. Access ends at expiry, and the server deletes the data within one hour.
Can I recover a lost job link?
There is no account or public task list through which to recover it. Save the complete private link and download results before expiry.
How should I cite DeepHVI?
Use the reference on the citation page, which provides the full bibliographic details, DOI, and BibTeX.