I apologize if the answer is obvious, but:
\nGiven only an RNA sequence, is there an easy way to tell which end is 3' and which end is 5'? Or, for the purpose of this competition, can we assume that every RNA sequence (training, test set, future test set, everything) always consistently starts from one end (either always 3', or always 5')?
\nWikipedia has the following. But can someone please confirm that the data given to us adhere to this convention?
\n\n\nBy convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing.
\n
Definition of easy: No supervised learning please. I'm looking for a knowledge-based way that requires zero training data.
", "messages": [ { "id": "3157766", "postDate": "03/23/2025 19:22:03", "content": "I apologize if the answer is obvious, but:
\nGiven only an RNA sequence, is there an easy way to tell which end is 3' and which end is 5'? Or, for the purpose of this competition, can we assume that every RNA sequence (training, test set, future test set, everything) always consistently starts from one end (either always 3', or always 5')?
\nWikipedia has the following. But can someone please confirm that the data given to us adhere to this convention?
\n\n\nBy convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing.
\n
Definition of easy: No supervised learning please. I'm looking for a knowledge-based way that requires zero training data.
", "rawMarkdown": "I apologize if the answer is obvious, but:\n\nGiven only an RNA sequence, is there an **easy** way to tell which end is 3' and which end is 5'? Or, for the purpose of this competition, can we assume that every RNA sequence (training, test set, future test set, everything) always consistently starts from one end (either always 3', or always 5')?\n\nWikipedia has the following. But can someone please confirm that the data given to us adhere to this convention?\n\n>By convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing. \n\nDefinition of **easy**: No supervised learning please. I'm looking for a knowledge-based way that requires zero training data.", "votes": null }, { "id": "3157782", "postDate": "03/23/2025 19:48:09", "content": "RNA sequence always starts from 5'-end and ends with 3'-end
", "rawMarkdown": "RNA sequence always starts from 5'-end and ends with 3'-end", "votes": null }, { "id": "3158646", "postDate": "03/24/2025 18:32:07", "content": "In case you are interested, 5' end contains a free phosphate group and 3' end contains a free hydroxyl group. In a whole nucleic acid chain, there is a single free phosphate (5') and a single free hydroxyl group (3'), as all others are involved in making a phospho-diester backbone. In simple terms, all others are involved in making side-rails of a ladder that holds nucleotides. As to why they are labeled 5' and 3', it has to do with carbons in a ribose molecule (a total of five) to which these groups are attached.
\n\n\nBy convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing.
\n
Let's say you see a double-stranded DNA molecule written like this:
\n\nIn this case the top strand would be written in a 5'->3' direction, and the bottom strand, which is anti-parallel, would be written in a 3'->5' direction. That's the exception mentioned in the above statement.
\n'-AGAACTATG'\n'-TCTTGATAC'\nSince RNA doesn't have the second strand - its double-stranded regions are created intramolecularly - it is always written 5'->3'.
", "rawMarkdown": "In case you are interested, 5' end contains a free phosphate group and 3' end contains a free hydroxyl group. In a whole nucleic acid chain, there is a single free phosphate (5') and a single free hydroxyl group (3'), as all others are involved in making a phospho-diester backbone. In simple terms, all others are involved in making side-rails of a ladder that holds nucleotides. As to why they are labeled 5' and 3', it has to do with carbons in a ribose molecule (a total of five) to which these groups are attached.\n\n> By convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing.\n\nLet's say you see a double-stranded DNA molecule written like this:\n\n AGAACTATG\n TCTTGATAC\n\nIn this case the top strand would be written in a 5'->3' direction, and the bottom strand, which is anti-parallel, would be written in a 3'->5' direction. That's the exception mentioned in the above statement.\n\n 5'-AGAACTATG-3'\n 3'-TCTTGATAC-5'\n\nSince RNA doesn't have the second strand - its double-stranded regions are created intramolecularly - it is always written 5'->3'.", "votes": null }, { "id": "3158854", "postDate": "03/25/2025 01:50:58", "content": "Thank you for the explanation. If two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3', is it safe from a molecular biology point of view that these two RNA sequences will fold to identical or nearly identical shape in 3D?
\nFor example: 5'-UUACC…GGUAA-3' vs 3'-UUACC…GGUAA-5'
\nPut it another way, would the free (and presumably exposed) phosphate/hydroxyl groups behave any differently, due to the fact that they're free, as far as intramolecular interaction (maybe via hydrogen bond/Van der Walls/what-have-you) is concerned?
\nEven for those non-free phosphate/hydroxyl groups within the sequence, what if we reverse the 5'-3' direction but otherwise keep the sequence the same?
\nThank you!
", "rawMarkdown": "Thank you for the explanation. If two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3', is it safe from a molecular biology point of view that these two RNA sequences will fold to identical or nearly identical shape in 3D?\n\nFor example: 5'-UUACC...GGUAA-3' vs 3'-UUACC...GGUAA-5'\n\nPut it another way, would the free (and presumably exposed) phosphate/hydroxyl groups behave any differently, due to the fact that they're free, as far as intramolecular interaction (maybe via hydrogen bond/Van der Walls/what-have-you) is concerned?\n\nEven for those non-free phosphate/hydroxyl groups within the sequence, what if we reverse the 5'-3' direction but otherwise keep the sequence the same?\n\nThank you!", "votes": null }, { "id": "3158866", "postDate": "03/25/2025 02:11:10", "content": "\n\nIf two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3'
\n
There are no RNAs starting at 3' end. I guess what you mean is a complementary RNA, but it would also start at its own 5' end. I don't think non-palindromic complementary RNAs fold the same way, but I have no formal proof for that. In a special case of palindromic molecules, they might.
\nFree phosphates and hydroxyl groups have negligible effect (most likely zero effect) on RNA folding. They could affect RNA reactivity and its lifetime, but we don't need to worry about that as it has no bearing on folding.
", "rawMarkdown": "> If two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3'\n\nThere are no RNAs starting at 3' end. I guess what you mean is a complementary RNA, but it would also start at its own 5' end. I don't think non-palindromic complementary RNAs fold the same way, but I have no formal proof for that. In a special case of palindromic molecules, they might.\n\nFree phosphates and hydroxyl groups have negligible effect (most likely zero effect) on RNA folding. They could affect RNA reactivity and its lifetime, but we don't need to worry about that as it has no bearing on folding.", "votes": null }, { "id": "3160073", "postDate": "03/26/2025 10:41:21", "content": "This may be off-topic for this competition but I'm guessing whether we put data into the model by the order of 5'-3' or 3'-5' doesn't matter
", "rawMarkdown": "This may be off-topic for this competition but I'm guessing whether we put data into the model by the order of 5'-3' or 3'-5' doesn't matter", "votes": null }, { "id": "3160392", "postDate": "03/26/2025 17:50:53", "content": "\n\nThis may be off-topic for this competition
\n
The 5'-3' (a)symmetry has pretty non-trivial relevance for this competition even from a purely machine learning perspective without regard to molecular biology or chemistry.
\nI know it's not obvious, but it does.
", "rawMarkdown": ">This may be off-topic for this competition\n\nThe 5'-3' (a)symmetry has pretty non-trivial relevance for this competition even from a purely machine learning perspective without regard to molecular biology or chemistry.\n\nI know it's not obvious, but it does.", "votes": null }, { "id": "3160588", "postDate": "03/27/2025 00:15:21", "content": "I know the basics of attention mechanisms, but I finetuned Ribonanzanet with reversed sequences+original sequences,and the scores didn't change(original sequence only:0.179 ori+reversed:0.181) so at least on finetuning level directions don't give substantial effect.
\nI'm guessing this might be because atoms in sequences are placed mostly to minimize the potential, and contexts have weaker effect on atom positions than potential
Thank you for your reply, and looking forward to your reply
\nI'm a bigginer at kaggle, discussion
\nlike this really motivates me& helps me grasp models&data feature more
(1) Most (all?) natural long RNA sequences data we have don't have both ground truths of themselves and their reversed counterparts available.
\n(2) If two very long, non-palindromic RNA sequences are reverses of each other, would they fold sufficiently differently (factoring out intrinsic uncertainties and variations)?
\nIf the answer to (2) is yes, then (1) and the fact that current models tend to produce the same/similar prediction when you reverse the sequence (as you observed) could imply studying how long sequence reversal affects folding could be exploited to improve current models.
\nBut of course, it's impossible to tell thanks to (1), short of using expensive chemistry equipment.
\nOr is it? 😀
", "rawMarkdown": "(1) Most (all?) natural long RNA sequences data we have don't have *both* ground truths of themselves and their reversed counterparts available.\n\n(2) If two very long, non-palindromic RNA sequences are reverses of each other, would they fold sufficiently differently (factoring out intrinsic uncertainties and variations)?\n\nIf the answer to (2) is yes, then (1) and the fact that current models tend to produce the same/similar prediction when you reverse the sequence (as you observed) could imply studying how long sequence reversal affects folding could be exploited to improve current models.\n\nBut of course, it's impossible to tell thanks to (1), short of using expensive chemistry equipment.\n\nOr is it? 😀", "votes": null }, { "id": "3161413", "postDate": "03/27/2025 21:58:32", "content": "\n\n(1) Most (all?) natural long RNA sequences data we have don't have both ground truths of themselves and their reversed counterparts available.
\n
None of them have both ground truths, because in biological terms there is only one ground truth. That's the one that will be structurally characterized and used in this competition. Complementary sequences of these RNAs are never studied for a reason: they are functionally irrelevant.
\nThere is no point really in discussing complementary sequences any further, as I don't think they will help the modeling in any way.
", "rawMarkdown": "> (1) Most (all?) natural long RNA sequences data we have don't have both ground truths of themselves and their reversed counterparts available.\n\nNone of them have both ground truths, because in biological terms there is only one ground truth. That's the one that will be structurally characterized and used in this competition. Complementary sequences of these RNAs are never studied for a reason: they are functionally irrelevant.\n\nThere is no point really in discussing complementary sequences any further, as I don't think they will help the modeling in any way.", "votes": null } ], "comments": [ { "id": 3157782, "author_name": "eugenebaulin", "author_url": "", "post_date": "03/23/2025 19:48:09", "content": "RNA sequence always starts from 5'-end and ends with 3'-end
", "votes": null, "replies": [] }, { "id": 3158646, "author_name": "tilii7", "author_url": "", "post_date": "03/24/2025 18:32:07", "content": "In case you are interested, 5' end contains a free phosphate group and 3' end contains a free hydroxyl group. In a whole nucleic acid chain, there is a single free phosphate (5') and a single free hydroxyl group (3'), as all others are involved in making a phospho-diester backbone. In simple terms, all others are involved in making side-rails of a ladder that holds nucleotides. As to why they are labeled 5' and 3', it has to do with carbons in a ribose molecule (a total of five) to which these groups are attached.
\n\n\nBy convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing.
\n
Let's say you see a double-stranded DNA molecule written like this:
\n\nIn this case the top strand would be written in a 5'->3' direction, and the bottom strand, which is anti-parallel, would be written in a 3'->5' direction. That's the exception mentioned in the above statement.
\n'-AGAACTATG'\n'-TCTTGATAC'\nSince RNA doesn't have the second strand - its double-stranded regions are created intramolecularly - it is always written 5'->3'.
", "votes": null, "replies": [ { "id": 3158854, "author_name": "revealer", "author_url": "", "post_date": "03/25/2025 01:50:58", "content": "Thank you for the explanation. If two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3', is it safe from a molecular biology point of view that these two RNA sequences will fold to identical or nearly identical shape in 3D?
\nFor example: 5'-UUACC…GGUAA-3' vs 3'-UUACC…GGUAA-5'
\nPut it another way, would the free (and presumably exposed) phosphate/hydroxyl groups behave any differently, due to the fact that they're free, as far as intramolecular interaction (maybe via hydrogen bond/Van der Walls/what-have-you) is concerned?
\nEven for those non-free phosphate/hydroxyl groups within the sequence, what if we reverse the 5'-3' direction but otherwise keep the sequence the same?
\nThank you!
", "votes": null, "replies": [ { "id": 3158866, "author_name": "tilii7", "author_url": "", "post_date": "03/25/2025 02:11:10", "content": "\n\nIf two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3'
\n
There are no RNAs starting at 3' end. I guess what you mean is a complementary RNA, but it would also start at its own 5' end. I don't think non-palindromic complementary RNAs fold the same way, but I have no formal proof for that. In a special case of palindromic molecules, they might.
\nFree phosphates and hydroxyl groups have negligible effect (most likely zero effect) on RNA folding. They could affect RNA reactivity and its lifetime, but we don't need to worry about that as it has no bearing on folding.
", "votes": null, "replies": [] } ] } ] }, { "id": 3160073, "author_name": "nanacat0520", "author_url": "", "post_date": "03/26/2025 10:41:21", "content": "This may be off-topic for this competition but I'm guessing whether we put data into the model by the order of 5'-3' or 3'-5' doesn't matter
", "votes": null, "replies": [ { "id": 3160392, "author_name": "revealer", "author_url": "", "post_date": "03/26/2025 17:50:53", "content": "\n\nThis may be off-topic for this competition
\n
The 5'-3' (a)symmetry has pretty non-trivial relevance for this competition even from a purely machine learning perspective without regard to molecular biology or chemistry.
\nI know it's not obvious, but it does.
", "votes": null, "replies": [ { "id": 3160588, "author_name": "nanacat0520", "author_url": "", "post_date": "03/27/2025 00:15:21", "content": "I know the basics of attention mechanisms, but I finetuned Ribonanzanet with reversed sequences+original sequences,and the scores didn't change(original sequence only:0.179 ori+reversed:0.181) so at least on finetuning level directions don't give substantial effect.
\nI'm guessing this might be because atoms in sequences are placed mostly to minimize the potential, and contexts have weaker effect on atom positions than potential
Thank you for your reply, and looking forward to your reply
\nI'm a bigginer at kaggle, discussion
\nlike this really motivates me& helps me grasp models&data feature more
(1) Most (all?) natural long RNA sequences data we have don't have both ground truths of themselves and their reversed counterparts available.
\n(2) If two very long, non-palindromic RNA sequences are reverses of each other, would they fold sufficiently differently (factoring out intrinsic uncertainties and variations)?
\nIf the answer to (2) is yes, then (1) and the fact that current models tend to produce the same/similar prediction when you reverse the sequence (as you observed) could imply studying how long sequence reversal affects folding could be exploited to improve current models.
\nBut of course, it's impossible to tell thanks to (1), short of using expensive chemistry equipment.
\nOr is it? 😀
", "votes": null, "replies": [ { "id": 3161413, "author_name": "tilii7", "author_url": "", "post_date": "03/27/2025 21:58:32", "content": "\n\n(1) Most (all?) natural long RNA sequences data we have don't have both ground truths of themselves and their reversed counterparts available.
\n
None of them have both ground truths, because in biological terms there is only one ground truth. That's the one that will be structurally characterized and used in this competition. Complementary sequences of these RNAs are never studied for a reason: they are functionally irrelevant.
\nThere is no point really in discussing complementary sequences any further, as I don't think they will help the modeling in any way.
", "votes": null, "replies": [] } ] } ] } ] } ] } ], "raw_markdown_by_id": { "3157766": "I apologize if the answer is obvious, but:\n\nGiven only an RNA sequence, is there an **easy** way to tell which end is 3' and which end is 5'? Or, for the purpose of this competition, can we assume that every RNA sequence (training, test set, future test set, everything) always consistently starts from one end (either always 3', or always 5')?\n\nWikipedia has the following. But can someone please confirm that the data given to us adhere to this convention?\n\n>By convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing. \n\nDefinition of **easy**: No supervised learning please. I'm looking for a knowledge-based way that requires zero training data.", "3157782": "RNA sequence always starts from 5'-end and ends with 3'-end", "3158646": "In case you are interested, 5' end contains a free phosphate group and 3' end contains a free hydroxyl group. In a whole nucleic acid chain, there is a single free phosphate (5') and a single free hydroxyl group (3'), as all others are involved in making a phospho-diester backbone. In simple terms, all others are involved in making side-rails of a ladder that holds nucleotides. As to why they are labeled 5' and 3', it has to do with carbons in a ribose molecule (a total of five) to which these groups are attached.\n\n> By convention, single strands of DNA and RNA sequences are written in a 5′-to-3′ direction except as needed to illustrate the pattern of base pairing.\n\nLet's say you see a double-stranded DNA molecule written like this:\n\n AGAACTATG\n TCTTGATAC\n\nIn this case the top strand would be written in a 5'->3' direction, and the bottom strand, which is anti-parallel, would be written in a 3'->5' direction. That's the exception mentioned in the above statement.\n\n 5'-AGAACTATG-3'\n 3'-TCTTGATAC-5'\n\nSince RNA doesn't have the second strand - its double-stranded regions are created intramolecularly - it is always written 5'->3'.", "3158854": "Thank you for the explanation. If two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3', is it safe from a molecular biology point of view that these two RNA sequences will fold to identical or nearly identical shape in 3D?\n\nFor example: 5'-UUACC...GGUAA-3' vs 3'-UUACC...GGUAA-5'\n\nPut it another way, would the free (and presumably exposed) phosphate/hydroxyl groups behave any differently, due to the fact that they're free, as far as intramolecular interaction (maybe via hydrogen bond/Van der Walls/what-have-you) is concerned?\n\nEven for those non-free phosphate/hydroxyl groups within the sequence, what if we reverse the 5'-3' direction but otherwise keep the sequence the same?\n\nThank you!", "3158866": "> If two long, non-palindromic RNA sequences are completely identical, but one starts out 5' and the other starts out 3'\n\nThere are no RNAs starting at 3' end. I guess what you mean is a complementary RNA, but it would also start at its own 5' end. I don't think non-palindromic complementary RNAs fold the same way, but I have no formal proof for that. In a special case of palindromic molecules, they might.\n\nFree phosphates and hydroxyl groups have negligible effect (most likely zero effect) on RNA folding. They could affect RNA reactivity and its lifetime, but we don't need to worry about that as it has no bearing on folding.", "3160073": "This may be off-topic for this competition but I'm guessing whether we put data into the model by the order of 5'-3' or 3'-5' doesn't matter", "3160392": ">This may be off-topic for this competition\n\nThe 5'-3' (a)symmetry has pretty non-trivial relevance for this competition even from a purely machine learning perspective without regard to molecular biology or chemistry.\n\nI know it's not obvious, but it does.", "3160588": "I know the basics of attention mechanisms, but I finetuned Ribonanzanet with reversed sequences+original sequences,and the scores didn't change(original sequence only:0.179 ori+reversed:0.181) so at least on finetuning level directions don't give substantial effect.\nI'm guessing this might be because atoms in sequences are placed mostly to minimize the potential, and contexts have weaker effect on atom positions than potential\n\n\nThank you for your reply, and looking forward to your reply\nI'm a bigginer at kaggle, discussion\nlike this really motivates me& helps me grasp models&data feature more", "3161377": "(1) Most (all?) natural long RNA sequences data we have don't have *both* ground truths of themselves and their reversed counterparts available.\n\n(2) If two very long, non-palindromic RNA sequences are reverses of each other, would they fold sufficiently differently (factoring out intrinsic uncertainties and variations)?\n\nIf the answer to (2) is yes, then (1) and the fact that current models tend to produce the same/similar prediction when you reverse the sequence (as you observed) could imply studying how long sequence reversal affects folding could be exploited to improve current models.\n\nBut of course, it's impossible to tell thanks to (1), short of using expensive chemistry equipment.\n\nOr is it? 😀", "3161413": "> (1) Most (all?) natural long RNA sequences data we have don't have both ground truths of themselves and their reversed counterparts available.\n\nNone of them have both ground truths, because in biological terms there is only one ground truth. That's the one that will be structurally characterized and used in this competition. Complementary sequences of these RNAs are never studied for a reason: they are functionally irrelevant.\n\nThere is no point really in discussing complementary sequences any further, as I don't think they will help the modeling in any way." }, "source": "meta" }