SciForge: Skill for Codex

AlphaFold3 C550 Inference Platform

Submit, monitor, and download AlphaFold3 protein structure predictions on the C550 inference cluster.

Always distinguish three states in reports: (1) task submission accepted, (2) AF3 inference reached a terminal state, and (3) CIF coordinates were retrieved and analyzed. A queued or running task is not molecular evidence, and a completed task without CIF retrieval is execution-level evidence only.

Quick Start

# 1. Auto-generate input from a simple sequence
bash .codex/skills/alphafold3/scripts/alphafold3_submit.sh \
  --sequence "MQIFVKTLTGKTITLEVEPSDTIENVKAKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQKESTLHLVLRLRGG" \
  --name "My_Protein" \
  --output-dir ./outputs/alphafold3/my_run

# 0. Preflight control-plane and data-plane readiness before evidence runs
bash .codex/skills/alphafold3/scripts/alphafold3_submit.sh --preflight

# 2. Or use a pre-prepared JSON file
bash .codex/skills/alphafold3/scripts/alphafold3_submit.sh \
  --json-input ./my_input.json \
  --output-dir ./outputs/alphafold3/my_run

# 3. Or submit a PVC directory that already contains AF3 JSON files
bash .codex/skills/alphafold3/scripts/alphafold3_submit.sh \
  --input-dir /data/input/my_batch \
  --output-dir ./outputs/alphafold3/my_run

# 4. Or submit one JSON file that is already present on the PVC
bash .codex/skills/alphafold3/scripts/alphafold3_submit.sh \
  --remote-json /data/input/my_batch/my_input.json \
  --output-dir ./outputs/alphafold3/my_run

The script handles upload → submit → poll → download automatically for local JSON/sequence inputs. With --input-dir, it skips upload and submits the existing PVC directory directly through the HTTP control plane. With --remote-json, it submits one existing PVC JSON file through the HTTP control plane and avoids whole-directory failure when another JSON in the directory is invalid.

Input Format

Minimal working AlphaFold3 JSON:

{
  "name": "My_Protein",
  "dialect": "alphafold3",
  "version": 1,
  "modelSeeds": [1],
  "sequences": [
    {
      "protein": {
        "id": "A",
        "sequence": "MQIFVKTLTG...",
        "unpairedMsa": "",
        "pairedMsa": "",
        "templates": []
      }
    }
  ]
}