After reading the \"Data\" section of this competition, I understand we are given training data that covers RNA Sequence from length 115 to 206.
\nThe test data is on RNA Sequence from length 207 to 457.
\nI am novice in RNA. I would like to know is the underlying model theoretically believed to be same which predicts 115 to 206 seq as well as which predicts 207 to 457 ?
\nWill the model be able to learn the generalized features of seq 207 to 457 (which is expected from the participant's model) from the training data of seq 115 to 206?
", "messages": [ { "id": "2513031", "postDate": "11/05/2023 05:46:47", "content": "After reading the \"Data\" section of this competition, I understand we are given training data that covers RNA Sequence from length 115 to 206.
\nThe test data is on RNA Sequence from length 207 to 457.
\nI am novice in RNA. I would like to know is the underlying model theoretically believed to be same which predicts 115 to 206 seq as well as which predicts 207 to 457 ?
\nWill the model be able to learn the generalized features of seq 207 to 457 (which is expected from the participant's model) from the training data of seq 115 to 206?
", "rawMarkdown": "After reading the \"Data\" section of this competition, I understand we are given training data that covers RNA Sequence from length 115 to 206. \n\nThe test data is on RNA Sequence from length 207 to 457. \n\nI am novice in RNA. I would like to know is the underlying model theoretically believed to be same which predicts 115 to 206 seq as well as which predicts 207 to 457 ? \n\nWill the model be able to learn the generalized features of seq 207 to 457 (which is expected from the participant's model) from the training data of seq 115 to 206?", "votes": null }, { "id": "2513582", "postDate": "11/05/2023 14:43:48", "content": "The original 2D (secondary structure) prediction methods used nearest neighbor thermodynamic calculations to find the minimum free energy structure of an RNA molecule, and naturally those thermodynamic energies still exist in RNA molecules. Part of what we are testing with this challenge is whether the patterns recognized at length 115 to 206 will generalize to longer lengths. We believe they can!
\nRNA molecules fold in 2D before making 3D (tertiary) contacts, generally. A model that can predict the 3D contacts is the ultimate goal as demonstrated by Shujun's pinned post. Is there any RNA biology relevant to the longer lengths beyond that? Not that I know of. One aspect of RNA that might be helpful to understand is pseudoknots.
", "rawMarkdown": "The original 2D (secondary structure) prediction methods used nearest neighbor thermodynamic calculations to find the minimum free energy structure of an RNA molecule, and naturally those thermodynamic energies still exist in RNA molecules. Part of what we are testing with this challenge is whether the patterns recognized at length 115 to 206 will generalize to longer lengths. We believe they can!\n\nRNA molecules fold in 2D before making 3D (tertiary) contacts, generally. A model that can predict the 3D contacts is the ultimate goal as demonstrated by Shujun's pinned post. Is there any RNA biology relevant to the longer lengths beyond that? Not that I know of. One aspect of RNA that might be helpful to understand is [pseudoknots](https://eternagame.org/collections/11366215).", "votes": null }, { "id": "2513720", "postDate": "11/05/2023 16:43:22", "content": "Judging by this post, yes. It is possible to generalize, at least to some degree.
", "rawMarkdown": "Judging by [this post](https://www.kaggle.com/competitions/stanford-ribonanza-rna-folding/discussion/444653), yes. It is possible to generalize, at least to some degree.", "votes": null }, { "id": "2513738", "postDate": "11/05/2023 17:01:08", "content": "Thank you for the explanation and link to pseudoknots.
", "rawMarkdown": "Thank you for the explanation and link to pseudoknots.", "votes": null }, { "id": "2513748", "postDate": "11/05/2023 17:11:01", "content": "This post you shared is to check your model generalized or not.
\nMy question is more on the lines that if there are some intrinsic differences in sequence from length 115 to 206 vs sequence 207 to 457; then irrespective of the model architecture, since the training data is from lengths 115 to 206 the model will only learn features from 115 to 206, it won't learn features(that intrinsically differentiate) 207 to 457.
", "rawMarkdown": "This post you shared is to check your model generalized or not. \n\nMy question is more on the lines that if there are some intrinsic differences in sequence from length 115 to 206 vs sequence 207 to 457; then irrespective of the model architecture, since the training data is from lengths 115 to 206 the model will only learn features from 115 to 206, it won't learn features(that intrinsically differentiate) 207 to 457.", "votes": null } ], "comments": [ { "id": 2513582, "author_name": "digitalembrace", "author_url": "", "post_date": "11/05/2023 14:43:48", "content": "The original 2D (secondary structure) prediction methods used nearest neighbor thermodynamic calculations to find the minimum free energy structure of an RNA molecule, and naturally those thermodynamic energies still exist in RNA molecules. Part of what we are testing with this challenge is whether the patterns recognized at length 115 to 206 will generalize to longer lengths. We believe they can!
\nRNA molecules fold in 2D before making 3D (tertiary) contacts, generally. A model that can predict the 3D contacts is the ultimate goal as demonstrated by Shujun's pinned post. Is there any RNA biology relevant to the longer lengths beyond that? Not that I know of. One aspect of RNA that might be helpful to understand is pseudoknots.
", "votes": null, "replies": [ { "id": 2513738, "author_name": "mink007", "author_url": "", "post_date": "11/05/2023 17:01:08", "content": "Thank you for the explanation and link to pseudoknots.
", "votes": null, "replies": [] } ] }, { "id": 2513720, "author_name": "shlomoron", "author_url": "", "post_date": "11/05/2023 16:43:22", "content": "Judging by this post, yes. It is possible to generalize, at least to some degree.
", "votes": null, "replies": [ { "id": 2513748, "author_name": "mink007", "author_url": "", "post_date": "11/05/2023 17:11:01", "content": "This post you shared is to check your model generalized or not.
\nMy question is more on the lines that if there are some intrinsic differences in sequence from length 115 to 206 vs sequence 207 to 457; then irrespective of the model architecture, since the training data is from lengths 115 to 206 the model will only learn features from 115 to 206, it won't learn features(that intrinsically differentiate) 207 to 457.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2513031": "After reading the \"Data\" section of this competition, I understand we are given training data that covers RNA Sequence from length 115 to 206. \n\nThe test data is on RNA Sequence from length 207 to 457. \n\nI am novice in RNA. I would like to know is the underlying model theoretically believed to be same which predicts 115 to 206 seq as well as which predicts 207 to 457 ? \n\nWill the model be able to learn the generalized features of seq 207 to 457 (which is expected from the participant's model) from the training data of seq 115 to 206?", "2513582": "The original 2D (secondary structure) prediction methods used nearest neighbor thermodynamic calculations to find the minimum free energy structure of an RNA molecule, and naturally those thermodynamic energies still exist in RNA molecules. Part of what we are testing with this challenge is whether the patterns recognized at length 115 to 206 will generalize to longer lengths. We believe they can!\n\nRNA molecules fold in 2D before making 3D (tertiary) contacts, generally. A model that can predict the 3D contacts is the ultimate goal as demonstrated by Shujun's pinned post. Is there any RNA biology relevant to the longer lengths beyond that? Not that I know of. One aspect of RNA that might be helpful to understand is [pseudoknots](https://eternagame.org/collections/11366215).", "2513720": "Judging by [this post](https://www.kaggle.com/competitions/stanford-ribonanza-rna-folding/discussion/444653), yes. It is possible to generalize, at least to some degree.", "2513738": "Thank you for the explanation and link to pseudoknots.", "2513748": "This post you shared is to check your model generalized or not. \n\nMy question is more on the lines that if there are some intrinsic differences in sequence from length 115 to 206 vs sequence 207 to 457; then irrespective of the model architecture, since the training data is from lengths 115 to 206 the model will only learn features from 115 to 206, it won't learn features(that intrinsically differentiate) 207 to 457." }, "source": "meta" }