Protein Design Competition Platform

A reusable platform for running protein design competitions using AlphaFold3 structure prediction. Relies on Netlify for deployment.

Quick Start for SeokLab Members

If you're in the SeokLab GitHub organization and want to run your own competition, the self-hosted runner on galaxy4 is already available for all seoklab repos.

Initial Setup

1. Create your repository under seoklab org
2. Clone this template and update paths (see Setup Guide below)
3. Replace target problems with your actual targets (see Replacing Target Problems)
4. Set up Netlify for the submission form
5. Enable GitHub Pages for the public site

---

Replacing Target Problems

The platform comes with example problems. Here's how to replace them with your actual competition targets.

Files to Modify

File	Purpose
docs/targets/config.json	Problem definitions (names, types, residue counts)
docs/targets/*.pdb	Reference structures for evaluation
docs/targets/*.a3m	Pre-computed MSA files (for binder targets)
.github/workflows/check_completion.yml	Problem-to-reference mapping for evaluation
docs/index.html	Problem descriptions shown to participants

Step 1: Prepare Reference Structures

For each problem, prepare a backbone-only PDB file:

# Extract backbone atoms (N, CA, C, O) from your reference structure
# Name convention: {problem_id}_bb.pdb or descriptive name

# For monomer problems:
grep -E "^ATOM.*(  N  |  CA |  C  |  O  )" full_structure.pdb > problem_1_bb.pdb

# For binder problems, include both target and binder chains:
# Chain A = participant's binder, Chain B = given target

Place all PDB files in docs/targets/.

Step 2: Update config.json

Edit docs/targets/config.json:

{
  "problems": [
    {
      "id": "problem_1",
      "name": "Problem 1 - Your Title",
      "description": "Description shown to participants",
      "target_file": "your_target_1.pdb",
      "residue_count": 50,
      "type": "monomer",
      "msa_mode": "none"
    },
    {
      "id": "problem_2",
      "name": "Problem 2 - Binder Design",
      "description": "Design a binder for target X",
      "type": "binder",
      "target_file": "binder_problem_x.pdb",
      "residue_count": 80,
      "target_sequence": "MKTAYIAK...",
      "target_msa_file": "/full/path/to/target_x.a3m",
      "participant_msa_mode": "none",
      "expected_binder_length": [50, 100]
    }
  ]
}

Problem Types:

• monomer: Participant designs a sequence to fold into the target structure
• binder: Participant designs a binder sequence for a given target protein

MSA Modes:

• none: No MSA search (faster, for de novo design)
• search: Run MSA search (default AF3 behavior)
• precomputed: Use pre-calculated MSA file

Step 3: Update Workflow Evaluation Mapping

Edit .github/workflows/check_completion.yml, find the PROBLEM_REFS section (~lines 224-230):

declare -A PROBLEM_REFS=(
  ["problem_1"]="your_target_1.pdb:monomer"
  ["problem_2"]="binder_problem_x.pdb:binder"
  ["problem_3"]="your_target_3.pdb:monomer"
  # Add/remove as needed
)

Format: ["problem_id"]="reference_file.pdb:problem_type"

Step 4: Update Submission Form

Edit docs/index.html to update problem descriptions shown to participants:

<div class="problem-section">
  <h3>Problem 1: Your Title</h3>
  <p>Description of what participants should design...</p>
  <p><strong>Target length:</strong> 50 residues</p>
</div>

Step 5: For Binder Problems - Prepare MSA

If your binder problem uses a pre-computed MSA for the target:

# Generate MSA using your preferred tool (e.g., jackhmmer, hhblits)
# Save as .a3m format
# Reference the full path in config.json target_msa_file field

Step 6: Clear Old Results (if any)

# Remove old evaluation results
rm -rf docs/results/*/

# Reset leaderboard
echo '{"problems": {}, "overall_rankings": [], "last_updated": ""}' > docs/leaderboard_data.json

Example: Current Demo Problems

Problem	Type	Reference	Residues
Problem 1	monomer	3v86_bb.pdb	27
Problem 2	monomer	4r80_bb.pdb	76
Problem 3	monomer	1qys_bb.pdb	91
Problem 4	monomer	6wi5_bb.pdb	92
Problem 5	binder	binder_problem_9bk5.pdb	78

---

Setup Guide

Step 1: Create Your Repository

git clone https://github.com/seoklab/design-test.git my-competition
cd my-competition
git remote set-url origin https://github.com/seoklab/my-competition.git

Step 2: Update Paths

Replace paths with your username in these files:

.github/workflows/process_submission.yml (~lines 15, 58):

SUBMISSIONS_BASE: /data/galaxy4/user/YOUR_USER/my-competition/submissions
QUEUE_DIR: /data/galaxy4/user/YOUR_USER/job_queue

.github/workflows/check_completion.yml (~lines 15-18):

SUBMISSIONS_BASE: /data/galaxy4/user/YOUR_USER/my-competition/submissions
PUBLIC_RESULTS: /data/galaxy4/user/YOUR_USER/my-competition/public_results
SITE_URL: https://your-site.netlify.app
ADMIN_EMAIL: your-email@example.com

Step 3: Create Directories on Galaxy4

mkdir -p /data/galaxy4/user/YOUR_USER/my-competition/submissions
mkdir -p /data/galaxy4/user/YOUR_USER/my-competition/public_results
mkdir -p /data/galaxy4/user/YOUR_USER/job_queue
chmod 777 /data/galaxy4/user/YOUR_USER/my-competition/submissions
chmod 777 /data/galaxy4/user/YOUR_USER/my-competition/public_results

Step 4: Set Up Job Queue Cron

Add to crontab on galaxy4 (crontab -e):

*/15 * * * * for f in /data/galaxy4/user/YOUR_USER/job_queue/*.sh; do [ -f "$f" ] && sbatch "$f" && mv "$f" "$f.done"; done

Step 5: Set Up Netlify

1. Create account at netlify.com
2. Add new site → Import an existing project → Connect to GitHub
3. Select your repository
4. Add environment variables in Site settings:
   • GITHUB_TOKEN: Personal access token with repo scope
   • GITHUB_OWNER: seoklab
   • GITHUB_REPO: Your repo name

Step 6: Update Form Submit URL

Edit docs/index.html:

const SUBMIT_URL = 'https://your-site.netlify.app/api/submit';

Step 7: Enable GitHub Pages

1. Settings → Pages
2. Source: Deploy from a branch
3. Branch: main, Folder: /docs

---

Architecture

User → Web Form → Netlify Function → GitHub Actions
       (GitHub Pages)                     ↓
                                   Process Submission
                                   (Self-hosted runner on galaxy4)
                                          ↓
                                   SLURM Job Queue → AlphaFold3
                                          ↓
                                   Check Completion (every minute)
                                          ↓
                                   Evaluate + Update Leaderboard
                                          ↓
                              Email + Viewer Link → User

File Structure

├── .github/workflows/
│   ├── process_submission.yml    # Handles new submissions
│   ├── check_completion.yml      # Checks jobs, evaluates, updates leaderboard
│   └── end_competition.yml       # Final evaluation workflow
├── docs/                         # Public site (GitHub Pages)
│   ├── index.html                # Submission form
│   ├── viewer.html               # Mol* structure viewer
│   ├── leaderboard.html          # Competition leaderboard
│   ├── leaderboard_data.json     # Leaderboard data
│   ├── targets/                  # Reference structures
│   │   ├── config.json           # Problem definitions
│   │   └── *.pdb                 # Reference PDB files
│   └── results/                  # Packaged results (per token)
├── netlify/functions/
│   └── submit.js                 # Form submission API
├── scripts/
│   ├── process_multi_submission.py  # Parse submission
│   ├── run_af3.py                   # Generate SLURM script
│   ├── package_results.py           # Package results with token
│   ├── evaluate_structure.py        # TMalign/lDDT evaluation
│   └── update_leaderboard.py        # Aggregate rankings
└── submissions/                  # (gitignored) On galaxy4

Evaluation Metrics

Monomer Problems

Metric	Description
BB-lDDT	Backbone local distance difference test (primary)
RMSD	Root mean square deviation after alignment
TM-score	Template modeling score
pTM	AlphaFold predicted TM-score
pLDDT	AlphaFold predicted lDDT

Binder Problems

Category	Metrics
Binder Only	BB-lDDT, RMSD, TM-score (binder chain vs reference)
Complex	iLDDT, TM-score, RMSD (full complex alignment)
AlphaFold3	ipTM, pTM, pLDDT

Maintenance

Enable/Disable Scheduled Checks

In .github/workflows/check_completion.yml:

on:
  schedule:
    - cron: '* * * * *'  # Comment out to disable
  workflow_dispatch:      # Keep for manual triggers

Manual Workflow Trigger

Actions → Check Job Completion → Run workflow

Netlify Auto-Deploy (Optional)

To auto-deploy when results are ready:

1. In Netlify: Site settings → Build hooks → Add hook
2. Add as GitHub secret: NETLIFY_BUILD_HOOK
3. Add step in check_completion.yml after commit:

- name: Trigger Netlify deploy
  if: steps.check.outputs.results_updated == 'true'
  run: curl -X POST -d {} ${{ secrets.NETLIFY_BUILD_HOOK }}

Troubleshooting

Issue	Solution
Git push rejected	git pull --rebase && git push
Permission denied	Run chmod 777 on submission directories
Form not submitting	Check Netlify env vars and function logs
No emails	Check sendmail config, may have delay (5-10 min normal)
Workflow not running	Check Actions tab, verify runner is online
Wrong evaluation scores	Ensure reference PDB residue numbering starts at 1

Credits

• AlphaFold3 - Structure prediction
• TMalign/USalign - Structure alignment
• PDBe-Molstar - 3D visualization
• SeokLab