![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F13967749%2F84afbcb35144d44f2474141b818d74d9%2F173074781d26e7bd97457169ff2e8885.png?generation=1747809839381953&alt=media\")
Insert trRosettaRNA results below 350 into the fifth position of ProteniX can increase the score to 0.381
Insert trRosettaRNA results below 350 into the fifth position of ProteniX can increase the score to 0.381
0.381 config: seed=42
\nusing torch.use_deterministic_algorithms(True) to make protenix deterministic:
\nhttps://www.kaggle.com/code/thekog/deterministic-protenix
There is a trick I forgot to post
\nyou can use op_score from pyrosetta to sort the result of protenix, to prevent the best output being replaced:
\nimport pyrosetta
\nfrom pyrosetta import pose_from_pdb, create_score_function
\nfrom pyrosetta.rosetta import core
pyrosetta.init(options='-mute all')
\nop_score = create_score_function('ref2015')
\nop_score.set_weight(core.scoring.atom_pair_constraint, 9.0)
\nop_score.set_weight(core.scoring.dihedral_constraint, 4.0)
\nop_score.set_weight(core.scoring.angle_constraint, 4.0)
\nop_score.set_weight(core.scoring.fa_rep, 9.0)
\nop_score.set_weight(core.scoring.rna_sugar_close, 9.0)
\nop_score.set_weight(core.scoring.fa_intra_rep, 9.0)
\nop_score.set_weight(core.scoring.rna_base_pair, 9.0)
\nop_score.set_weight(core.scoring.rna_base_stack, 9.0)
def protenix_to_pdb_and_df(coordinate, atom_array, pdb_path):
\n # Create a PandasPdb object
\n lines= []
\n for i, atom in enumerate(atom_array):
\n # Extract atom information
\n atom_name = atom.atom_name
\n res_name = atom.res_name
\n chain_id = atom.chain_id
\n res_id = atom.res_id
\n x, y, z = coordinate[i][0].item(), coordinate[i][1].item(), coordinate[i][2].item()
\n line = \n lines.append(line)\n\n (pdb_path, ) f:\n f.writelines(lines)\n\ndf = pd.DataFrame({\n : [atom.atom_name atom atom_array],\n : [atom.res_name atom atom_array],\n : [atom.chain_id atom atom_array],\n : [atom.res_id atom atom_array],\n : [coordinate[i][].item() i ((atom_array))],\n : [coordinate[i][].item() i ((atom_array))],\n : [coordinate[i][].item() i ((atom_array))],\n : [atom.element atom atom_array]\n})\n df\nbefore using op score, you need to convert prediction['coordinate'] to pdb file, using protenix_to_pdb_and_df()
", "rawMarkdown": "There is a trick I forgot to post\n\nyou can use op_score from pyrosetta to sort the result of protenix, to prevent the best output being replaced:\n\n\nimport pyrosetta\nfrom pyrosetta import pose_from_pdb, create_score_function\nfrom pyrosetta.rosetta import core\n\npyrosetta.init(options='-mute all') \n\nop_score = create_score_function('ref2015')\nop_score.set_weight(core.scoring.atom_pair_constraint, 9.0)\nop_score.set_weight(core.scoring.dihedral_constraint, 4.0)\nop_score.set_weight(core.scoring.angle_constraint, 4.0)\nop_score.set_weight(core.scoring.fa_rep, 9.0)\nop_score.set_weight(core.scoring.rna_sugar_close, 9.0)\nop_score.set_weight(core.scoring.fa_intra_rep, 9.0)\nop_score.set_weight(core.scoring.rna_base_pair, 9.0)\nop_score.set_weight(core.scoring.rna_base_stack, 9.0)\n\ndef protenix_to_pdb_and_df(coordinate, atom_array, pdb_path):\n # Create a PandasPdb object\n lines= []\n for i, atom in enumerate(atom_array):\n # Extract atom information\n atom_name = atom.atom_name\n res_name = atom.res_name\n chain_id = atom.chain_id\n res_id = atom.res_id\n x, y, z = coordinate[i][0].item(), coordinate[i][1].item(), coordinate[i][2].item()\n\n # Create PDB line\n line = f\"ATOM {i+1:>5} {atom_name:<4} {res_name:<3} {chain_id:<1}{res_id:>4} {x:>8.3f}{y:>8.3f}{z:>8.3f} 1.00 1.00 {atom.element:<2}\\n\"\n lines.append(line)\n # Write to PDB file\n with open(pdb_path, 'w') as f:\n f.writelines(lines)\n # Create DataFrame\n df = pd.DataFrame({\n 'atom_name': [atom.atom_name for atom in atom_array],\n 'res_name': [atom.res_name for atom in atom_array],\n 'chain_id': [atom.chain_id for atom in atom_array],\n 'res_id': [atom.res_id for atom in atom_array],\n 'x': [coordinate[i][0].item() for i in range(len(atom_array))],\n 'y': [coordinate[i][1].item() for i in range(len(atom_array))],\n 'z': [coordinate[i][2].item() for i in range(len(atom_array))],\n 'element': [atom.element for atom in atom_array]\n })\n return df\n\n\nbefore using op score, you need to convert prediction['coordinate'] to pdb file, using protenix_to_pdb_and_df()", "votes": 2 }, { "id": 3207674, "postDate": "2025-05-23T05:44:49.027Z", "content": "Is it necessary to use multiple sequence secondary structure optimization?
", "rawMarkdown": "Is it necessary to use multiple sequence secondary structure optimization?", "votes": 1 }, { "id": 3208133, "postDate": "2025-05-23T18:22:07.863Z", "content": "Not necessary
", "rawMarkdown": " Not necessary ", "votes": 1 }, { "id": 3209853, "postDate": "2025-05-26T12:12:26.380Z", "content": "Thx! I'm confused about the value of seed, which equal to 42, do you suggest that I change something else to test whether it will have an impact on the score?
", "rawMarkdown": "Thx! I'm confused about the value of seed, which equal to 42, do you suggest that I change something else to test whether it will have an impact on the score?", "votes": 1 } ] } ] }, { "id": 3206398, "postDate": "2025-05-21T08:48:42.067Z", "content": "Good trick, Though it might be a bit tricky for folks to the competition. Best of luck to any Kagglers reading this! 🚀
", "rawMarkdown": "Good trick, Though it might be a bit tricky for folks to the competition. Best of luck to any Kagglers reading this! 🚀", "votes": 1 }, { "id": 3206313, "postDate": "2025-05-21T06:44:32.290Z", "content": "https://www.kaggle.com/code/thekog/trrosettarna-runable
", "rawMarkdown": "https://www.kaggle.com/code/thekog/trrosettarna-runable", "votes": 1 }, { "id": 3207673, "postDate": "2025-05-23T05:43:53.227Z", "content": "Very useful hints! Thanks a lot!
", "rawMarkdown": "Very useful hints! Thanks a lot!" } ], "comments": [ { "id": 3206417, "author_name": "", "author_url": "", "post_date": "2025-05-21T09:23:20.040000", "content": "0.381 config: seed=42
\nusing torch.use_deterministic_algorithms(True) to make protenix deterministic:
\nhttps://www.kaggle.com/code/thekog/deterministic-protenix
There is a trick I forgot to post
\nyou can use op_score from pyrosetta to sort the result of protenix, to prevent the best output being replaced:
\nimport pyrosetta
\nfrom pyrosetta import pose_from_pdb, create_score_function
\nfrom pyrosetta.rosetta import core
pyrosetta.init(options='-mute all')
\nop_score = create_score_function('ref2015')
\nop_score.set_weight(core.scoring.atom_pair_constraint, 9.0)
\nop_score.set_weight(core.scoring.dihedral_constraint, 4.0)
\nop_score.set_weight(core.scoring.angle_constraint, 4.0)
\nop_score.set_weight(core.scoring.fa_rep, 9.0)
\nop_score.set_weight(core.scoring.rna_sugar_close, 9.0)
\nop_score.set_weight(core.scoring.fa_intra_rep, 9.0)
\nop_score.set_weight(core.scoring.rna_base_pair, 9.0)
\nop_score.set_weight(core.scoring.rna_base_stack, 9.0)
def protenix_to_pdb_and_df(coordinate, atom_array, pdb_path):
\n # Create a PandasPdb object
\n lines= []
\n for i, atom in enumerate(atom_array):
\n # Extract atom information
\n atom_name = atom.atom_name
\n res_name = atom.res_name
\n chain_id = atom.chain_id
\n res_id = atom.res_id
\n x, y, z = coordinate[i][0].item(), coordinate[i][1].item(), coordinate[i][2].item()
\n line = \n lines.append(line)\n\n (pdb_path, ) f:\n f.writelines(lines)\n\ndf = pd.DataFrame({\n : [atom.atom_name atom atom_array],\n : [atom.res_name atom atom_array],\n : [atom.chain_id atom atom_array],\n : [atom.res_id atom atom_array],\n : [coordinate[i][].item() i ((atom_array))],\n : [coordinate[i][].item() i ((atom_array))],\n : [coordinate[i][].item() i ((atom_array))],\n : [atom.element atom atom_array]\n})\n df\nbefore using op score, you need to convert prediction['coordinate'] to pdb file, using protenix_to_pdb_and_df()
", "votes": 2, "replies": [] }, { "id": 3207674, "author_name": "", "author_url": "", "post_date": "2025-05-23T05:44:49.027000", "content": "Is it necessary to use multiple sequence secondary structure optimization?
", "votes": 1, "replies": [] }, { "id": 3208133, "author_name": "Zhongyuan Zhang", "author_url": "", "post_date": "2025-05-23T18:22:07.863000", "content": "Not necessary
", "votes": 1, "replies": [] }, { "id": 3209853, "author_name": "", "author_url": "", "post_date": "2025-05-26T12:12:26.380000", "content": "Thx! I'm confused about the value of seed, which equal to 42, do you suggest that I change something else to test whether it will have an impact on the score?
", "votes": 1, "replies": [] } ] } ] }, { "id": 3206398, "author_name": "Tan_T64", "author_url": "", "post_date": "2025-05-21T08:48:42.067000", "content": "Good trick, Though it might be a bit tricky for folks to the competition. Best of luck to any Kagglers reading this! 🚀
", "votes": 1, "replies": [] }, { "id": 3206313, "author_name": "", "author_url": "", "post_date": "2025-05-21T06:44:32.290000", "content": "https://www.kaggle.com/code/thekog/trrosettarna-runable
", "votes": 1, "replies": [] }, { "id": 3207673, "author_name": "", "author_url": "", "post_date": "2025-05-23T05:43:53.227000", "content": "Very useful hints! Thanks a lot!
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3206312": "Insert trRosettaRNA results below 350 into the fifth position of ProteniX can increase the score to 0.381", "3206417": "0.381 config: seed=42\nusing torch.use_deterministic_algorithms(True) to make protenix deterministic:\nhttps://www.kaggle.com/code/thekog/deterministic-protenix", "3206398": "Good trick, Though it might be a bit tricky for folks to the competition. Best of luck to any Kagglers reading this! 🚀", "3206313": "https://www.kaggle.com/code/thekog/trrosettarna-runable", "3207673": "Very useful hints! Thanks a lot!" } }