![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8722753%2F667dd29e99f49c4233e8eb20f415288f%2Foutput.png?generation=1743074153282247&alt=media\")
I'm wondering if there are specific rules in the PDB for handling mirror-symmetric structures.
\nAs you know, when only the 1D (sequence) information is given and a model is required to predict the 3D structure, there are always at least two mirror-symmetric possibilities—unless the molecule is achiral (i.e., it matches its own mirror image).
\nFrom what I understand, USalign does not consider mirror-image structures to be equivalent (though I'm not entirely sure about this).
\nIf that's the case, wouldn't we need to submit both mirror-image structures for each sequence to ensure a match? That would reduce our submission capacity to 2.5 meaningful predictions per target.
\nIs my understanding correct? Or is there a specific convention or strategy in the field to deal with this issue?
", "messages": [ { "id": "3160809", "postDate": "03/27/2025 07:00:58", "content": "I'm wondering if there are specific rules in the PDB for handling mirror-symmetric structures.
\nAs you know, when only the 1D (sequence) information is given and a model is required to predict the 3D structure, there are always at least two mirror-symmetric possibilities—unless the molecule is achiral (i.e., it matches its own mirror image).
\nFrom what I understand, USalign does not consider mirror-image structures to be equivalent (though I'm not entirely sure about this).
\nIf that's the case, wouldn't we need to submit both mirror-image structures for each sequence to ensure a match? That would reduce our submission capacity to 2.5 meaningful predictions per target.
\nIs my understanding correct? Or is there a specific convention or strategy in the field to deal with this issue?
", "rawMarkdown": "I'm wondering if there are specific rules in the PDB for handling mirror-symmetric structures.\n\nAs you know, when only the 1D (sequence) information is given and a model is required to predict the 3D structure, there are always at least two mirror-symmetric possibilities—unless the molecule is achiral (i.e., it matches its own mirror image).\n\nFrom what I understand, USalign does *not* consider mirror-image structures to be equivalent (though I'm not entirely sure about this). \n\nIf that's the case, wouldn't we need to submit both mirror-image structures for each sequence to ensure a match? That would reduce our submission capacity to 2.5 meaningful predictions per target.\n\nIs my understanding correct? Or is there a specific convention or strategy in the field to deal with this issue?", "votes": null }, { "id": "3160956", "postDate": "03/27/2025 10:47:51", "content": "The rule is the one obtained experimentally. But you can observe BAT coordinates and take your own conclusions. Most RNA's helical structure is right-handed.
\nTorsion angles ABCD:
\n
The RNA molecule's building block the nucleotide has a chiral D-ribose (a ribose that is not equivalent to its mirror image). So the mirror image of any RNA structure contains L-ribose, which does not occur naturally in nucleotides.
\nIt is a bit unclear to me what you mean by
\n\n\nAs you know, when only the 1D (sequence) information is given and a model is required to predict the 3D structure, there are always at least two mirror-symmetric possibilities
\n
Do you mean something like this?
\nU-A
\nG-C
\nC-G
\nMirrored:
\nA-U
\nC-G
\nG-C
These are not real mirror images, they can be rotated to obtain one another. If they were mirror images one would have to contain naturally not occurring nucleotides (with L-ribose).
\nOr do you mean for example right and left handed helices?
\nSince the building blocks are chiral a right handed helix is different from the mirror image of a left a handed helix. They are not real mirror images of each other (one would have to contain L-ribose for that)!
\nThis also means that from the sequence you can 'easily' tell a left and a right handed helix.
\n\nSo the mirror image of any RNA structure contains L-ribose, which does not occur naturally in nucleotides.
\n
I missed that point — thank you for pointing it out.
\n\n\nThis also means that from the sequence you can 'easily' tell a left and a right handed helix.
\n
It’s true that a model might output a structure with L-ribose if it naively minimizes potential energy without considering chirality. However, I believe this is already handled in most cases, and such mirror structures can be filtered out in principle, as you stated.
", "rawMarkdown": "> So the mirror image of any RNA structure contains L-ribose, which does not occur naturally in nucleotides.\n\nI missed that point — thank you for pointing it out.\n\n> This also means that from the sequence you can 'easily' tell a left and a right handed helix.\n\nIt’s true that a model might output a structure with L-ribose if it naively minimizes potential energy without considering chirality. However, I believe this is already handled in most cases, and such mirror structures can be filtered out in principle, as you stated.", "votes": null }, { "id": "3173806", "postDate": "04/08/2025 11:09:38", "content": "Left handed helix doesn't mean that it contains L-ribose.
\nLeft and right handed helices both contain D-ribose .
\nImagine it like this:
\nYou are stacking the monomers on top of each other with a little twist each time (resulting in a helix). You can either twist every new monomer a bit clockwise (left handed helix) or a bit anticlockwise (right handed helix).
\nThese two options are not rotationally equivalent.
\nThey are also not mirror images of each other. They look like mirror images without closer inspection, but they actually aren't, because they both contain D-ribose.
You can imagine that one of these stackings is more favorable. For example if every monomer has a 'bulky' group stick out on the top right, than it is more favorable to build the chain with the anticlockwise twist, giving space to the bulky part. That is why I said it is easy to predict left and right handed helices, they are fundamentally different. As the other commenter pointed out, most RNA helices are right handed.
", "rawMarkdown": "Left handed helix doesn't mean that it contains L-ribose.\nLeft and right handed helices both contain D-ribose . \nImagine it like this:\nYou are stacking the monomers on top of each other with a little twist each time (resulting in a helix). You can either twist every new monomer a bit clockwise (left handed helix) or a bit anticlockwise (right handed helix).\nThese two options are not rotationally equivalent.\nThey are also not mirror images of each other. They look like mirror images without closer inspection, but they actually aren't, because they both contain D-ribose.\n\nYou can imagine that one of these stackings is more favorable. For example if every monomer has a 'bulky' group stick out on the top right, than it is more favorable to build the chain with the anticlockwise twist, giving space to the bulky part. That is why I said it is easy to predict left and right handed helices, they are fundamentally different. As the other commenter pointed out, most RNA helices are right handed.", "votes": null } ], "comments": [ { "id": 3160956, "author_name": "sacuscreed", "author_url": "", "post_date": "03/27/2025 10:47:51", "content": "The rule is the one obtained experimentally. But you can observe BAT coordinates and take your own conclusions. Most RNA's helical structure is right-handed.
\nTorsion angles ABCD:
\n
The RNA molecule's building block the nucleotide has a chiral D-ribose (a ribose that is not equivalent to its mirror image). So the mirror image of any RNA structure contains L-ribose, which does not occur naturally in nucleotides.
\nIt is a bit unclear to me what you mean by
\n\n\nAs you know, when only the 1D (sequence) information is given and a model is required to predict the 3D structure, there are always at least two mirror-symmetric possibilities
\n
Do you mean something like this?
\nU-A
\nG-C
\nC-G
\nMirrored:
\nA-U
\nC-G
\nG-C
These are not real mirror images, they can be rotated to obtain one another. If they were mirror images one would have to contain naturally not occurring nucleotides (with L-ribose).
\nOr do you mean for example right and left handed helices?
\nSince the building blocks are chiral a right handed helix is different from the mirror image of a left a handed helix. They are not real mirror images of each other (one would have to contain L-ribose for that)!
\nThis also means that from the sequence you can 'easily' tell a left and a right handed helix.
\n\nSo the mirror image of any RNA structure contains L-ribose, which does not occur naturally in nucleotides.
\n
I missed that point — thank you for pointing it out.
\n\n\nThis also means that from the sequence you can 'easily' tell a left and a right handed helix.
\n
It’s true that a model might output a structure with L-ribose if it naively minimizes potential energy without considering chirality. However, I believe this is already handled in most cases, and such mirror structures can be filtered out in principle, as you stated.
", "votes": null, "replies": [ { "id": 3173806, "author_name": "gyulamaloveczky4", "author_url": "", "post_date": "04/08/2025 11:09:38", "content": "Left handed helix doesn't mean that it contains L-ribose.
\nLeft and right handed helices both contain D-ribose .
\nImagine it like this:
\nYou are stacking the monomers on top of each other with a little twist each time (resulting in a helix). You can either twist every new monomer a bit clockwise (left handed helix) or a bit anticlockwise (right handed helix).
\nThese two options are not rotationally equivalent.
\nThey are also not mirror images of each other. They look like mirror images without closer inspection, but they actually aren't, because they both contain D-ribose.
You can imagine that one of these stackings is more favorable. For example if every monomer has a 'bulky' group stick out on the top right, than it is more favorable to build the chain with the anticlockwise twist, giving space to the bulky part. That is why I said it is easy to predict left and right handed helices, they are fundamentally different. As the other commenter pointed out, most RNA helices are right handed.
", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3160809": "I'm wondering if there are specific rules in the PDB for handling mirror-symmetric structures.\n\nAs you know, when only the 1D (sequence) information is given and a model is required to predict the 3D structure, there are always at least two mirror-symmetric possibilities—unless the molecule is achiral (i.e., it matches its own mirror image).\n\nFrom what I understand, USalign does *not* consider mirror-image structures to be equivalent (though I'm not entirely sure about this). \n\nIf that's the case, wouldn't we need to submit both mirror-image structures for each sequence to ensure a match? That would reduce our submission capacity to 2.5 meaningful predictions per target.\n\nIs my understanding correct? Or is there a specific convention or strategy in the field to deal with this issue?", "3160956": "The rule is the one obtained experimentally. But you can observe [BAT](https://www.kaggle.com/code/sacuscreed/xyz-to-bat-and-back) coordinates and take your own conclusions. Most RNA's helical structure is right-handed.\n\n**Torsion angles ABCD:**\n\n", "3173195": "The RNA molecule's building block the nucleotide has a chiral D-ribose (a ribose that is not equivalent to its mirror image). So the mirror image of any RNA structure contains L-ribose, which does not occur naturally in nucleotides.\n\nIt is a bit unclear to me what you mean by \n>As you know, when only the 1D (sequence) information is given and a model is required to predict the 3D structure, there are always at least two mirror-symmetric possibilities\n\nDo you mean something like this?\n\nU-A\nG-C\nC-G\nMirrored:\nA-U\nC-G\nG-C\n\nThese are not real mirror images, they can be rotated to obtain one another. If they were mirror images one would have to contain naturally not occurring nucleotides (with L-ribose).\n\nOr do you mean for example right and left handed helices?\nSince the building blocks are chiral a right handed helix is different from the mirror image of a left a handed helix. They are not real mirror images of each other (one would have to contain L-ribose for that)!\nThis also means that from the sequence you can 'easily' tell a left and a right handed helix.", "3173464": "> So the mirror image of any RNA structure contains L-ribose, which does not occur naturally in nucleotides.\n\nI missed that point — thank you for pointing it out.\n\n> This also means that from the sequence you can 'easily' tell a left and a right handed helix.\n\nIt’s true that a model might output a structure with L-ribose if it naively minimizes potential energy without considering chirality. However, I believe this is already handled in most cases, and such mirror structures can be filtered out in principle, as you stated.", "3173806": "Left handed helix doesn't mean that it contains L-ribose.\nLeft and right handed helices both contain D-ribose . \nImagine it like this:\nYou are stacking the monomers on top of each other with a little twist each time (resulting in a helix). You can either twist every new monomer a bit clockwise (left handed helix) or a bit anticlockwise (right handed helix).\nThese two options are not rotationally equivalent.\nThey are also not mirror images of each other. They look like mirror images without closer inspection, but they actually aren't, because they both contain D-ribose.\n\nYou can imagine that one of these stackings is more favorable. For example if every monomer has a 'bulky' group stick out on the top right, than it is more favorable to build the chain with the anticlockwise twist, giving space to the bulky part. That is why I said it is easy to predict left and right handed helices, they are fundamentally different. As the other commenter pointed out, most RNA helices are right handed." }, "source": "meta" }