![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2262596%2F6f04c32a6ce47d2bdcfd18afe8d36c47%2FScreenshot%202023-09-08%20174832.png?generation=1694188152363916&alt=media\")
If anyone is thinking of Graph Neural Networks, it could theoretically work. Basically, we represent the RNA structure as a graph.
\nNodes: Each nucleotide in the RNA sequence can be represented as a node. The node features can include the nucleotide type (A, U, C, G), along with any other engineered features (e.g., reactivity scores, thermodynamic properties, etc.)
\nEdges: These can represent the interactions between nucleotides. These interactions could be physical bonds, or they could be potential interactions based on proximity or some other criteria.
\nNode & Edge Updates: At each layer of the GNN, nodes and edges are updated based on the features of their neighbors and potentially global features as well.
\nReadout Layer: After several layers of updates, the features of the nodes and/or edges are aggregated to make the final prediction.
", "messages": [ { "id": "2429404", "postDate": "09/08/2023 15:27:15", "content": "If anyone is thinking of Graph Neural Networks, it could theoretically work. Basically, we represent the RNA structure as a graph.
\nNodes: Each nucleotide in the RNA sequence can be represented as a node. The node features can include the nucleotide type (A, U, C, G), along with any other engineered features (e.g., reactivity scores, thermodynamic properties, etc.)
\nEdges: These can represent the interactions between nucleotides. These interactions could be physical bonds, or they could be potential interactions based on proximity or some other criteria.
\nNode & Edge Updates: At each layer of the GNN, nodes and edges are updated based on the features of their neighbors and potentially global features as well.
\nReadout Layer: After several layers of updates, the features of the nodes and/or edges are aggregated to make the final prediction.
", "rawMarkdown": "If anyone is thinking of Graph Neural Networks, it could theoretically work. Basically, we represent the RNA structure as a graph.\n\nNodes: Each nucleotide in the RNA sequence can be represented as a node. The node features can include the nucleotide type (A, U, C, G), along with any other engineered features (e.g., reactivity scores, thermodynamic properties, etc.)\n\nEdges: These can represent the interactions between nucleotides. These interactions could be physical bonds, or they could be potential interactions based on proximity or some other criteria.\n\nNode & Edge Updates: At each layer of the GNN, nodes and edges are updated based on the features of their neighbors and potentially global features as well.\n\nReadout Layer: After several layers of updates, the features of the nodes and/or edges are aggregated to make the final prediction.", "votes": null }, { "id": "2429446", "postDate": "09/08/2023 15:51:25", "content": "GNN were in previous Kaggle comp on RNA (https://www.kaggle.com/c/stanford-covid-vaccine) , so may be there is a chance for them to be used here also. See e.g.
\nhttps://youtu.be/sp3kZwKKYfw?si=1vYXsI-THY-XXcXz&t=1275:
\n
\"Graph Transformer\": https://www.kaggle.com/competitions/stanford-covid-vaccine/discussion/183865
\n\"GCN simpler\" https://www.kaggle.com/jameschapman19/openvaccine-gcn
", "rawMarkdown": "GNN were in previous Kaggle comp on RNA (https://www.kaggle.com/c/stanford-covid-vaccine) , so may be there is a chance for them to be used here also. See e.g.\n\nhttps://youtu.be/sp3kZwKKYfw?si=1vYXsI-THY-XXcXz&t=1275:\n\n\n\n\"Graph Transformer\": https://www.kaggle.com/competitions/stanford-covid-vaccine/discussion/183865\n\n\"GCN simpler\" https://www.kaggle.com/jameschapman19/openvaccine-gcn", "votes": null }, { "id": "2431211", "postDate": "09/09/2023 22:29:40", "content": "There is a directly related work: Physics-aware Graph Neural Network for Accurate RNA 3D Structure Prediction https://arxiv.org/abs/2210.16392
", "rawMarkdown": "There is a directly related work: Physics-aware Graph Neural Network for Accurate RNA 3D Structure Prediction https://arxiv.org/abs/2210.16392", "votes": null } ], "comments": [ { "id": 2429446, "author_name": "alexandervc", "author_url": "", "post_date": "09/08/2023 15:51:25", "content": "GNN were in previous Kaggle comp on RNA (https://www.kaggle.com/c/stanford-covid-vaccine) , so may be there is a chance for them to be used here also. See e.g.
\nhttps://youtu.be/sp3kZwKKYfw?si=1vYXsI-THY-XXcXz&t=1275:
\n
\"Graph Transformer\": https://www.kaggle.com/competitions/stanford-covid-vaccine/discussion/183865
\n\"GCN simpler\" https://www.kaggle.com/jameschapman19/openvaccine-gcn
", "votes": null, "replies": [] }, { "id": 2431211, "author_name": "cheny03", "author_url": "", "post_date": "09/09/2023 22:29:40", "content": "There is a directly related work: Physics-aware Graph Neural Network for Accurate RNA 3D Structure Prediction https://arxiv.org/abs/2210.16392
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2429404": "If anyone is thinking of Graph Neural Networks, it could theoretically work. Basically, we represent the RNA structure as a graph.\n\nNodes: Each nucleotide in the RNA sequence can be represented as a node. The node features can include the nucleotide type (A, U, C, G), along with any other engineered features (e.g., reactivity scores, thermodynamic properties, etc.)\n\nEdges: These can represent the interactions between nucleotides. These interactions could be physical bonds, or they could be potential interactions based on proximity or some other criteria.\n\nNode & Edge Updates: At each layer of the GNN, nodes and edges are updated based on the features of their neighbors and potentially global features as well.\n\nReadout Layer: After several layers of updates, the features of the nodes and/or edges are aggregated to make the final prediction.", "2429446": "GNN were in previous Kaggle comp on RNA (https://www.kaggle.com/c/stanford-covid-vaccine) , so may be there is a chance for them to be used here also. See e.g.\n\nhttps://youtu.be/sp3kZwKKYfw?si=1vYXsI-THY-XXcXz&t=1275:\n\n\n\n\"Graph Transformer\": https://www.kaggle.com/competitions/stanford-covid-vaccine/discussion/183865\n\n\"GCN simpler\" https://www.kaggle.com/jameschapman19/openvaccine-gcn", "2431211": "There is a directly related work: Physics-aware Graph Neural Network for Accurate RNA 3D Structure Prediction https://arxiv.org/abs/2210.16392" }, "source": "meta" }