I am not a biologist, so I was wondering about the specific characteristics that make thyroglobulin and β-galactosidase more challenging to identify in cryo-EM images. Are these challenges primarily due to factors like size, shape, orientation, or other properties? I am thinking that perhaps because they are tiny, they appear very blurry in the 3D image.
", "messages": [ { "id": "3093848", "postDate": "01/11/2025 12:10:10", "content": "I am not a biologist, so I was wondering about the specific characteristics that make thyroglobulin and β-galactosidase more challenging to identify in cryo-EM images. Are these challenges primarily due to factors like size, shape, orientation, or other properties? I am thinking that perhaps because they are tiny, they appear very blurry in the 3D image.
", "rawMarkdown": "I am not a biologist, so I was wondering about the specific characteristics that make thyroglobulin and β-galactosidase more challenging to identify in cryo-EM images. Are these challenges primarily due to factors like size, shape, orientation, or other properties? I am thinking that perhaps because they are tiny, they appear very blurry in the 3D image.", "votes": null }, { "id": "3094364", "postDate": "01/12/2025 01:53:54", "content": "My suspicion is that part of the problem is that it's just very difficult to get well labelled data for these particles. (Which probably not coincidentally is one of the applications of these models.)
", "rawMarkdown": "My suspicion is that part of the problem is that it's just very difficult to get well labelled data for these particles. (Which probably not coincidentally is one of the applications of these models.)", "votes": null }, { "id": "3095787", "postDate": "01/13/2025 17:47:25", "content": "Thanks for the reply. Do you mean that it label for the hard particles are not very accurate(mislabelled) or there are not enough label (class imbalance)?
", "rawMarkdown": "Thanks for the reply. Do you mean that it label for the hard particles are not very accurate(mislabelled) or there are not enough label (class imbalance)?", "votes": null }, { "id": "3095792", "postDate": "01/13/2025 17:55:40", "content": "Hello. There are multiple reasons for the difficulty, some are inherent (smaller things like THG and beta-galactosidase look like each other more often than larger things, so the confusion is higher with those), and some of difficulty is due to labeling. I can say that we are very confident in our labels for these particles (check out our paper on biorxiv), but we are not error-proof because it's simply impossible with the current state of technologies.
", "rawMarkdown": "Hello. There are multiple reasons for the difficulty, some are inherent (smaller things like THG and beta-galactosidase look like each other more often than larger things, so the confusion is higher with those), and some of difficulty is due to labeling. I can say that we are very confident in our labels for these particles (check out our paper on biorxiv), but we are not error-proof because it's simply impossible with the current state of technologies.", "votes": null }, { "id": "3096235", "postDate": "01/14/2025 06:45:31", "content": "@rezaparaan On second read, I meant \"perfectly labeled data\" instead of \"well labeled data\". I'm seeing my model pick up unlabeled particles that look almost identical to the labeled ones…of course, that doesn't necessarily mean that's what they are.
", "rawMarkdown": "rezaparaan On second read, I meant \"perfectly labeled data\" instead of \"well labeled data\". I'm seeing my model pick up unlabeled particles that look almost identical to the labeled ones...of course, that doesn't necessarily mean that's what they are.", "votes": null }, { "id": "3096268", "postDate": "01/14/2025 07:24:42", "content": "@successmoses take a look through this notebook. It should help you understand some of the difficulty:
\nhttps://www.kaggle.com/code/davidlist/full-catalog-of-labeled-particles
My suspicion is that part of the problem is that it's just very difficult to get well labelled data for these particles. (Which probably not coincidentally is one of the applications of these models.)
", "votes": null, "replies": [ { "id": 3095787, "author_name": "successmoses", "author_url": "", "post_date": "01/13/2025 17:47:25", "content": "Thanks for the reply. Do you mean that it label for the hard particles are not very accurate(mislabelled) or there are not enough label (class imbalance)?
", "votes": null, "replies": [ { "id": 3095792, "author_name": "rezaparaan", "author_url": "", "post_date": "01/13/2025 17:55:40", "content": "Hello. There are multiple reasons for the difficulty, some are inherent (smaller things like THG and beta-galactosidase look like each other more often than larger things, so the confusion is higher with those), and some of difficulty is due to labeling. I can say that we are very confident in our labels for these particles (check out our paper on biorxiv), but we are not error-proof because it's simply impossible with the current state of technologies.
", "votes": null, "replies": [ { "id": 3096235, "author_name": "davidlist", "author_url": "", "post_date": "01/14/2025 06:45:31", "content": "@rezaparaan On second read, I meant \"perfectly labeled data\" instead of \"well labeled data\". I'm seeing my model pick up unlabeled particles that look almost identical to the labeled ones…of course, that doesn't necessarily mean that's what they are.
", "votes": null, "replies": [] } ] } ] } ] }, { "id": 3096268, "author_name": "davidlist", "author_url": "", "post_date": "01/14/2025 07:24:42", "content": "@successmoses take a look through this notebook. It should help you understand some of the difficulty:
\nhttps://www.kaggle.com/code/davidlist/full-catalog-of-labeled-particles