I apologize if is not an appropiate question but since I couldn't find public information about US-align I wonder if is \"singular\" alignment. I assume that this method gets the best possible alignment. I'm not sure how to implement a reasonable metric even a proper loss approach. There is infinite xyz possible sets for each rigid structure. I've considered to train with some kind of predefined internal coordinates independent to translations and rotations, but then how to chose the proper orientation to submit?
\nEDIT: That sentence was just wrong.
", "messages": [ { "id": "3139471", "postDate": "03/03/2025 14:49:04", "content": "I apologize if is not an appropiate question but since I couldn't find public information about US-align I wonder if is \"singular\" alignment. I assume that this method gets the best possible alignment. I'm not sure how to implement a reasonable metric even a proper loss approach. There is infinite xyz possible sets for each rigid structure. I've considered to train with some kind of predefined internal coordinates independent to translations and rotations, but then how to chose the proper orientation to submit?
\nEDIT: That sentence was just wrong.
", "rawMarkdown": "I apologize if is not an appropiate question but since I couldn't find public information about US-align I wonder if is \"singular\" alignment. ~~I mean, in the simplest approach when you rotate a structure by SVD decomposition there is 6x4 = 24 different orientations that are diagonal, 24x24 = 576 possible alignments.~~ I assume that this method gets the best possible alignment. I'm not sure how to implement a reasonable metric even a proper loss approach. There is infinite xyz possible sets for each rigid structure. I've considered to train with some kind of predefined internal coordinates independent to translations and rotations, but then how to chose the proper orientation to submit?\n\nEDIT: That sentence was just wrong.", "votes": null }, { "id": "3139595", "postDate": "03/03/2025 17:24:48", "content": "Molecules are rotated and translated in 3D space to minimize the Root Mean Square Deviation (RMSD). RMSD is a good proxy for the TM-score and they are inversely correlated.
\nThe orientation you submit is irrelevant, as the coordinate space is arbitrary. Coordinates will always be manipulated to create the best alignment with target structures.
", "rawMarkdown": "Molecules are rotated and translated in 3D space to minimize the [**Root Mean Square Deviation**](https://en.wikipedia.org/wiki/Root_mean_square_deviation) (RMSD). RMSD is a good proxy for the TM-score and they are inversely correlated.\n\nThe orientation you submit is irrelevant, as the coordinate space is arbitrary. Coordinates will always be manipulated to create the best alignment with target structures.", "votes": null }, { "id": "3139615", "postDate": "03/03/2025 17:44:04", "content": "\"Coordinates will always be manipulated to create the best alignment with target structures\" Thanks. I wanted to be sure about that.
", "rawMarkdown": "\"Coordinates will always be manipulated to create the best alignment with target structures\" Thanks. I wanted to be sure about that.", "votes": null }, { "id": "3142113", "postDate": "03/06/2025 05:31:28", "content": "Will be granulizing the USalign TM-score metric to multiple tree based ML algos using the starter features in the following notebook which already look promising for each prediction point set:
\nhttps://www.kaggle.com/code/jazivxt/tilt-or-fold
", "rawMarkdown": "Will be granulizing the USalign TM-score metric to multiple tree based ML algos using the starter features in the following notebook which already look promising for each prediction point set:\n\nhttps://www.kaggle.com/code/jazivxt/tilt-or-fold", "votes": null } ], "comments": [ { "id": 3139595, "author_name": "tilii7", "author_url": "", "post_date": "03/03/2025 17:24:48", "content": "Molecules are rotated and translated in 3D space to minimize the Root Mean Square Deviation (RMSD). RMSD is a good proxy for the TM-score and they are inversely correlated.
\nThe orientation you submit is irrelevant, as the coordinate space is arbitrary. Coordinates will always be manipulated to create the best alignment with target structures.
", "votes": null, "replies": [ { "id": 3139615, "author_name": "sacuscreed", "author_url": "", "post_date": "03/03/2025 17:44:04", "content": "\"Coordinates will always be manipulated to create the best alignment with target structures\" Thanks. I wanted to be sure about that.
", "votes": null, "replies": [] } ] }, { "id": 3142113, "author_name": "jazivxt", "author_url": "", "post_date": "03/06/2025 05:31:28", "content": "Will be granulizing the USalign TM-score metric to multiple tree based ML algos using the starter features in the following notebook which already look promising for each prediction point set:
\n", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "3139471": "I apologize if is not an appropiate question but since I couldn't find public information about US-align I wonder if is \"singular\" alignment. ~~I mean, in the simplest approach when you rotate a structure by SVD decomposition there is 6x4 = 24 different orientations that are diagonal, 24x24 = 576 possible alignments.~~ I assume that this method gets the best possible alignment. I'm not sure how to implement a reasonable metric even a proper loss approach. There is infinite xyz possible sets for each rigid structure. I've considered to train with some kind of predefined internal coordinates independent to translations and rotations, but then how to chose the proper orientation to submit?\n\nEDIT: That sentence was just wrong.", "3139595": "Molecules are rotated and translated in 3D space to minimize the [**Root Mean Square Deviation**](https://en.wikipedia.org/wiki/Root_mean_square_deviation) (RMSD). RMSD is a good proxy for the TM-score and they are inversely correlated.\n\nThe orientation you submit is irrelevant, as the coordinate space is arbitrary. Coordinates will always be manipulated to create the best alignment with target structures.", "3139615": "\"Coordinates will always be manipulated to create the best alignment with target structures\" Thanks. I wanted to be sure about that.", "3142113": "Will be granulizing the USalign TM-score metric to multiple tree based ML algos using the starter features in the following notebook which already look promising for each prediction point set:\n\nhttps://www.kaggle.com/code/jazivxt/tilt-or-fold" }, "source": "meta" }