feat(ProteinDesign): add FixedBackboneDesign task

benchmarks/ProteinDesign/FixedBackboneDesign/README.md

@@ -0,0 +1,25 @@
+# FixedBackboneDesign
+
+## Quick Start
+
+```bash
+cd benchmarks/ProteinDesign/FixedBackboneDesign
+mkdir -p outputs
+python verification/evaluator.py prepare --raw-task data/raw_task.json --prepared-output outputs/prepared.pdb
+python scripts/init.py --prepared-input outputs/prepared.pdb --solution-output outputs/solution.pdb
+python verification/evaluator.py evaluate --native outputs/prepared.pdb --candidate outputs/solution.pdb --result-output outputs/result.json
+```
+
+## Background
+
+This task is inspired by Agent Rosetta (Teneggi et al., arXiv:2603.15952, ICML 2026), which demonstrates LLM-driven protein design using Rosetta. This benchmark adapts the fixed-backbone sequence design problem as a standalone engineering optimization task.
+
+## Unified Benchmark ID
+
+`ProteinDesign/FixedBackboneDesign`
+
+## Environment
+
+```bash
+docker pull rosettacommons/rosetta:serial
+```

benchmarks/ProteinDesign/FixedBackboneDesign/README_zh-CN.md

@@ -0,0 +1,81 @@
+# FixedBackboneDesign - 固定骨架蛋白质序列设计
+
+## 一句话解释
+
+给定一个蛋白质的骨架结构（backbone），设计出最适合这个骨架的氨基酸序列，使 Rosetta 能量得分最低。
+
+## 任务描述
+
+本任务基于 Agent Rosetta 论文 (arXiv:2603.15952) 的 Case A：固定骨架规范氨基酸序列设计。Agent 需要从初始 baseline 出发，通过迭代优化找到总能量更低的氨基酸序列。
+
+## 文件结构
+
+```
+FixedBackboneDesign/
+├── Task.md                      # 任务详情（英文）
+├── Task_zh-CN.md                # 任务详情（中文）
+├── README.md                    # 本文件（导航）
+├── scripts/
+│   └── init.py                  # [可编辑] baseline 解法（含 EVOLVE-BLOCK）
+├── data/
+│   └── raw_task.json            # 任务配置
+├── references/
+│   ├── constants.json           # Rosetta 能量函数参数
+│   └── petrobind/               # PDB 结构文件
+├── frontier_eval/               # unified task 元数据
+├── verification/
+│   ├── evaluator.py             # [核心] PyRosetta 评分入口
+│   ├── requirements.txt
+│   └── docker/
+│       └── Dockerfile
+└── baseline/                    # [可选] 参考实现存档
+```
+
+## 运行方式
+
+### 本地三步运行
+
+```bash
+cd benchmarks/ProteinDesign/FixedBackboneDesign
+mkdir -p outputs
+
+# 1. prepare（由 evaluator 完成数据准备）
+python verification/evaluator.py prepare \
+  --raw-task data/raw_task.json \
+  --prepared-output outputs/prepared.pdb
+
+# 2. 运行 baseline 生成设计结果
+python scripts/init.py \
+  --prepared-input outputs/prepared.pdb \
+  --solution-output outputs/solution.pdb
+
+# 3. 评估设计结果
+python verification/evaluator.py evaluate \
+  --native outputs/prepared.pdb \
+  --candidate outputs/solution.pdb \
+  --result-output outputs/result.json
+```
+
+### Unified task 方式
+
+```bash
+python -m frontier_eval \
+  task=unified \
+  task.benchmark=ProteinDesign/FixedBackboneDesign \
+  task.runtime.isolation_mode=docker \
+  task.runtime.docker_image=rosettacommons/rosetta:serial \
+  algorithm=openevolve \
+  algorithm.iterations=0
+```
+
+## 统一 benchmark ID
+
+- `ProteinDesign/FixedBackboneDesign`
+
+## 环境准备
+
+本任务需要 Rosetta 环境，推荐使用官方 Docker 镜像：
+
+```bash
+docker pull rosettacommons/rosetta:serial
+```

benchmarks/ProteinDesign/FixedBackboneDesign/Task.md

@@ -0,0 +1,37 @@
+# FixedBackboneDesign
+
+## Overview
+
+Design the optimal amino acid sequence for a given fixed protein backbone. This task corresponds to Case A (Fixed-Backbone Canonical Amino Acid Sequence Design) of the Agent Rosetta paper (arXiv:2603.15952, ICML 2026).
+
+## Input
+
+The candidate script (`scripts/init.py`) receives:
+
+- `--prepared-input`: Path to a PDB file with the target backbone and design position annotations
+- `--solution-output`: Path where the designed structure should be written
+
+## Output
+
+The candidate must output a PDB file at `--solution-output` containing:
+1. The original backbone coordinates (unchanged)
+2. Designed amino acid substitutions at specified positions
+3. Only standard 20 amino acids (no non-canonical residues)
+
+## Scoring
+
+The evaluator uses PyRosetta's `ref2015` score function to compute:
+
+- **total_energy**: Sum of all energy terms
+- **baseline_energy**: Energy of the native (starting) sequence
+- **improvement**: baseline_energy - total_energy (positive = better)
+- **combined_score**: Normalized improvement = improvement / |baseline_energy|
+
+Energy terms reported: `fa_atr`, `fa_rep`, `fa_sol`, `fa_elec`, `hbond_bb_sc`, `hbond_sc`, `p_aa_pp`, `ref`
+
+## Constraints
+
+- Only modify `scripts/init.py`
+- Do NOT modify backbone coordinates
+- Only use the 20 standard amino acids
+- Output must be a valid PDB file

benchmarks/ProteinDesign/FixedBackboneDesign/Task_zh-CN.md

@@ -0,0 +1,37 @@
+# FixedBackboneDesign - 固定骨架蛋白质序列设计
+
+## 概述
+
+给定一个固定的蛋白质骨架（backbone），设计出最稳定的氨基酸序列。本任务对应 Agent Rosetta 论文（arXiv:2603.15952, ICML 2026）的 Case A：固定骨架规范氨基酸序列设计。
+
+## 输入
+
+candidate 脚本（`scripts/init.py`）接收以下参数：
+
+- `--prepared-input`：包含目标骨架和设计位置标注的 PDB 文件路径
+- `--solution-output`：设计结果 PDB 的输出路径
+
+## 输出
+
+candidate 必须在 `--solution-output` 输出一个 PDB 文件，包含：
+1. 原始的骨架坐标（保持不变）
+2. 在指定位置替换为设计后的氨基酸
+3. 只能使用 20 种标准氨基酸（不允许非规范氨基酸）
+
+## 评分方式
+
+评估器使用 PyRosetta 的 `ref2015` 能量函数计算：
+
+- **total_energy**：所有能量项的加和
+- **baseline_energy**：初始序列的能量
+- **improvement**：baseline_energy - total_energy（正数表示改进）
+- **combined_score**：归一化改进值 = improvement / |baseline_energy|
+
+报告的能量项：`fa_atr`、`fa_rep`、`fa_sol`、`fa_elec`、`hbond_bb_sc`、`hbond_sc`、`p_aa_pp`、`ref`
+
+## 约束
+
+- 只能修改 `scripts/init.py`
+- 不得修改骨架坐标
+- 只能使用 20 种标准氨基酸
+- 输出必须是有效的 PDB 文件

benchmarks/ProteinDesign/FixedBackboneDesign/baseline/result_log.txt

@@ -0,0 +1,31 @@
+FixedBackboneDesign baseline (no additional indexes beyond initial)
+==============================================================
+Date: 2026-07-14
+Evaluator: verification/evaluator.py (evaluate() function)
+Candidate: scripts/init.py (baseline: single-round PackRotamers with default settings)
+Environment: Docker (rosettacommons/rosetta:serial), PyRosetta ref2015
+
+Results:
+--------
+valid:            1.0
+combined_score:   0.025
+native_energy:   -169.18
+total_energy:   -173.48
+improvement:      +4.30
+
+Energy terms:
+  fa_atr:      -377.73
+  fa_rep:       110.97
+  fa_sol:       245.04
+  fa_elec:     -133.95
+  hbond_bb_sc:   -4.62
+  hbond_sc:      -6.87
+  p_aa_pp:      -23.02
+  ref:           12.74
+
+Note:
+- The baseline uses a single-round PackRotamersMover with packer.task().
+- Non-design positions are prevented from repacking via prevent_repacking().
+- Constraint verification (_verify_constraints) passed: no backbone movement,
+  no unexpected mutations at non-design positions.
+- Random seed fixed to 42 for reproducibility.

benchmarks/ProteinDesign/FixedBackboneDesign/data/raw_task.json

@@ -0,0 +1,14 @@
+{
+  "task_name": "fixed_backbone_design_demo",
+  "description": "Fixed-backbone protein sequence design using PyRosetta ref2015 scoring",
+  "pdb_id": "1ubq",
+  "pdb_path": "references/petrobind/1ubq.pdb",
+  "design_positions": [2, 4, 6, 8, 10, 12, 14, 16],
+  "native_sequence": "MQIFVKTLTGKTITLEVEPSDTIENVKAKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQKESTLHLVLRLRGG",
+  "energy_function": "ref2015",
+  "task_config": {
+    "n_design_positions": 8,
+    "allow_design_of_native": true,
+    "description": "Design 8 core positions of ubiquitin (PDB 1UBQ)"
+  }
+}

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/agent_files.txt

@@ -0,0 +1,8 @@
+Task.md
+Task_zh-CN.md
+scripts/init.py
+verification/evaluator.py
+verification/requirements.txt
+data/raw_task.json
+references/constants.json
+references/petrobind/1ubq.pdb

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/artifact_files.txt

@@ -0,0 +1,3 @@
+metrics.json
+outputs/*.pdb
+outputs/*.meta.json

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/candidate_destination.txt

@@ -0,0 +1 @@
+scripts/init.py

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/constraints.txt

@@ -0,0 +1,6 @@
+1) Only modify `scripts/init.py` — it is the ONLY editable file.
+2) Keep CLI contract unchanged: the candidate must accept `--prepared-input` and `--solution-output`.
+3) Output must be a PDB file with designed amino acid substitutions.
+4) Only use the 20 standard amino acids (no non-canonical residues).
+5) Do NOT modify the backbone coordinates (fixed-backbone design).
+6) Optimize the Rosetta total_score (lower is better). Use the ref2015 score function.

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/copy_files.txt

@@ -0,0 +1,4 @@
+scripts
+data
+verification
+references

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/eval_command.txt

@@ -0,0 +1 @@
+{python} {benchmark}/verification/evaluator.py {candidate}

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/eval_cwd.txt

@@ -0,0 +1 @@
+.

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/initial_program.txt

@@ -0,0 +1 @@
+scripts/init.py

benchmarks/ProteinDesign/FixedBackboneDesign/frontier_eval/readonly_files.txt

@@ -0,0 +1,3 @@
+data
+verification
+references

benchmarks/ProteinDesign/FixedBackboneDesign/references/constants.json

@@ -0,0 +1,13 @@
+{
+  "energy_function": "ref2015",
+  "score_terms": {
+    "fa_atr": {"weight": 1.0, "description": "Lennard-Jones attractive (van der Waals)"},
+    "fa_rep": {"weight": 0.55, "description": "Lennard-Jones repulsive (steric clash)"},
+    "fa_sol": {"weight": 1.0, "description": "Lazaridis-Karplus solvation energy"},
+    "fa_elec": {"weight": 1.0, "description": "Coulomb electrostatic potential"},
+    "hbond_bb_sc": {"weight": 1.0, "description": "Backbone-sidechain hydrogen bonds"},
+    "hbond_sc": {"weight": 1.0, "description": "Sidechain-sidechain hydrogen bonds"},
+    "p_aa_pp": {"weight": 0.6, "description": "Ramachandran torsion preference"},
+    "ref": {"weight": 1.0, "description": "Per-residue reference energy"}
+  }
+}