Tencent ORI Protein Sequence Generation and Prediction

Protein Generation and Prediction Workflows

On the Tencent Health Genomics Platform, the ORI Collection (Nextflow) exposes several sub-apps for protein generation, signal peptide prediction, solubility prediction, thermostability prediction, and structure prediction. Direct CLI use is easy to misconfigure: the collection AppId cannot be passed directly as --public-app to omics run, Nextflow sub-apps still need a valid --nf-version, importing into a project can collide with an existing app name, and trial users may be limited by remaining days and daily run quotas.

Core Capabilities and Key Steps

The skill turns these constraints into a controlled execution path:
- Scope locking: It serves only ORI Collection (Nextflow) and keeps the five sub-app AppIds hard-coded, preventing accidental expansion into unrelated collections.
- Environment checks: It verifies CLI availability, then validates login state, region, project, environment, and storage bucket settings through whoami and config show.
- Sub-app matching: It maps user intent to Generate Protein, USMFold Predict, Solubility, Thermostability, or Signal Peptide by name, index, or functional description.
- Import protection: It runs a same-name check before submission and asks for a new name on conflict instead of deleting or overwriting existing project apps.
- Submission control: It displays the full command and parameter summary, waits for explicit confirmation, and re-checks trial quotas before every omics run.
- Input adaptation: Locally generated files, such as FASTA sequences, are uploaded to the bound COS bucket before being used as run parameters.
- Failure handling: It gives clear next steps for DUPLICATE_APP_NAME, MISSING_NF_VERSION, PARAM_MERGE_FAILED, authentication errors, and other CLI failures.

Boundaries and Notes

The skill is intended for ORI protein generation and prediction tasks on the genomics platform. It should not be used to launch unrelated public apps or project-owned apps; out-of-scope requests are declined and redirected to a broader task skill. It does not edit application code automatically, does not submit tasks without confirmation, and treats async failures as evidence-gathering and rerun scenarios rather than silent pipeline fixes.