![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F8722753%2Fa6075daeb6d562273692fff1312d6518%2Fnewplot.png?generation=1741919359168622&alt=media\")
I've been working on cartesian to internal (bond, angle and torsion) conversion and viceversa with no succes. No matter what I do the conversion back is not the same original structure, not translated or rotated, actually different. After too much days arguing with DeepSeek is that faisable? I mean, one specific cartesian sequence has a specific non shared bond/angle/torsion sequence that allow come back to same cartesian sequence? And if is doable, how? Thanks for any suggestion.
", "messages": [ { "id": "3149023", "postDate": "03/13/2025 18:06:46", "content": "I've been working on cartesian to internal (bond, angle and torsion) conversion and viceversa with no succes. No matter what I do the conversion back is not the same original structure, not translated or rotated, actually different. After too much days arguing with DeepSeek is that faisable? I mean, one specific cartesian sequence has a specific non shared bond/angle/torsion sequence that allow come back to same cartesian sequence? And if is doable, how? Thanks for any suggestion.
", "rawMarkdown": "I've been working on cartesian to internal (bond, angle and torsion) conversion and viceversa with no succes. No matter what I do the conversion back is not the same original structure, not translated or rotated, actually different. After too much days arguing with DeepSeek is that faisable? I mean, one specific cartesian sequence has a specific non shared bond/angle/torsion sequence that allow come back to same cartesian sequence? And if is doable, how? Thanks for any suggestion.", "votes": null }, { "id": "3149165", "postDate": "03/13/2025 21:47:47", "content": "I don't think this is doable with C1 coordinates alone. Multiple atoms per nucleotide are required to define torsion angles and distances properly.
\nIt is possible to reconstruct a protein backbone from C-alpha atoms, but that's because there are only two additional atoms needed. Their bond lengths are fairly fixed and there are only so many torsional angle combinations that are feasible when it comes to proteins.
", "rawMarkdown": "I don't think this is doable with C1 coordinates alone. Multiple atoms per nucleotide are required to define torsion angles and distances properly.\n\nIt is possible to reconstruct a protein backbone from C-alpha atoms, but that's because there are only two additional atoms needed. Their bond lengths are fairly fixed and there are only so many torsional angle combinations that are feasible when it comes to proteins.", "votes": null }, { "id": "3149267", "postDate": "03/14/2025 02:26:53", "content": "After some pain we've got the solution.
", "rawMarkdown": "After some pain we've got the [solution](https://www.kaggle.com/code/sacuscreed/xyz-to-bat-and-back).", "votes": null }, { "id": "3149269", "postDate": "03/14/2025 02:29:20", "content": "Done!
\n
EDIT: I've used the plot function that compares predictions vs gt to compare reconstructed vs original cartesian coordinates.
", "rawMarkdown": "Done!\n\n\n\nEDIT: I've used the plot function that compares predictions vs gt to compare reconstructed vs original cartesian coordinates.", "votes": null } ], "comments": [ { "id": 3149165, "author_name": "tilii7", "author_url": "", "post_date": "03/13/2025 21:47:47", "content": "I don't think this is doable with C1 coordinates alone. Multiple atoms per nucleotide are required to define torsion angles and distances properly.
\nIt is possible to reconstruct a protein backbone from C-alpha atoms, but that's because there are only two additional atoms needed. Their bond lengths are fairly fixed and there are only so many torsional angle combinations that are feasible when it comes to proteins.
", "votes": null, "replies": [ { "id": 3149267, "author_name": "sacuscreed", "author_url": "", "post_date": "03/14/2025 02:26:53", "content": "After some pain we've got the solution.
", "votes": null, "replies": [] } ] }, { "id": 3149269, "author_name": "sacuscreed", "author_url": "", "post_date": "03/14/2025 02:29:20", "content": "Done!
\n
EDIT: I've used the plot function that compares predictions vs gt to compare reconstructed vs original cartesian coordinates.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3149023": "I've been working on cartesian to internal (bond, angle and torsion) conversion and viceversa with no succes. No matter what I do the conversion back is not the same original structure, not translated or rotated, actually different. After too much days arguing with DeepSeek is that faisable? I mean, one specific cartesian sequence has a specific non shared bond/angle/torsion sequence that allow come back to same cartesian sequence? And if is doable, how? Thanks for any suggestion.", "3149165": "I don't think this is doable with C1 coordinates alone. Multiple atoms per nucleotide are required to define torsion angles and distances properly.\n\nIt is possible to reconstruct a protein backbone from C-alpha atoms, but that's because there are only two additional atoms needed. Their bond lengths are fairly fixed and there are only so many torsional angle combinations that are feasible when it comes to proteins.", "3149267": "After some pain we've got the [solution](https://www.kaggle.com/code/sacuscreed/xyz-to-bat-and-back).", "3149269": "Done!\n\n\n\nEDIT: I've used the plot function that compares predictions vs gt to compare reconstructed vs original cartesian coordinates." }, "source": "meta" }