What problem does it solve? Running a real antibody design pipeline (RFdiffusion backbone generation, ProteinMPNN sequence design, Protenix structure prediction, and epitope screening) on Ascend NPUs requires careful environment setup, correct config authoring, and disciplined job monitoring. This Skill orchestrates that entire workflow through the ScienceDiscovery Runner NPU Broker so jobs are submitted once, monitored to a terminal state, and their artifacts registered correctly. ## Core Features & Use Cases - Broker-only job lifecycle: Lists workloads, submits antibody.protenix.v1 jobs with a workspace-local config.json, checkpoints the job ID, and polls status/logs until the job reaches a terminal state. - Managed environment validation: Probes ScienceDiscovery scientific environments for pinned dependencies (mindspore 2.7.2, numpy 1.26.4, biotite 1.4.0, rdkit 2024.3.5) and creates or updates environments before submission. - Custom antigen support: Accepts user-provided target antigen PDBs, antibody framework PDBs, and chain-labelled hotspot epitopes (e.g. [B45,B46,B49]) with strict validation of run names, NPU lists, and diffuser schedules. - Use Case: A researcher uploads a custom antigen PDB and antibody framework PDB, specifies epitope residues on chain B, and requests 4 designs across NPUs 0-3; the Skill prepares the config, submits the Broker job, monitors each stage, and returns declared Project Artifacts with screening results. ## Quick Start Run the Protenix antibody pipeline with my uploaded target antigen PDB and framework PDB, using hotspots B45,B46,B49, 4 designs on NPUs 0,1,2,3, and monitor the job until it finishes.
