What I assume to be simulated data - train_dataset1 - has clone sizes that are very different from real clones. the largest clone apears twice. Why is that? Is it not supposed to be repertoire data?
", "messages": [ { "id": "3311695", "postDate": "11/05/2025 15:22:54", "content": "What I assume to be simulated data - train_dataset1 - has clone sizes that are very different from real clones. the largest clone apears twice. Why is that? Is it not supposed to be repertoire data?
", "rawMarkdown": "What I assume to be simulated data - train_dataset1 - has clone sizes that are very different from real clones. the largest clone apears twice. Why is that? Is it not supposed to be repertoire data?", "votes": null }, { "id": "3312058", "postDate": "11/06/2025 10:47:49", "content": "Good observation! While the simulation of receptor sequence signals can also lean on insights from studies of antigen binding of individual receptors, there is no similar external basis to lean on regarding the clonal expansion of disease-relevant cells - in particular for the interaction between sequence signals and clonal expansion. To avoid any methods either exploiting unrealistically simulated clonal expansion patterns or getting thrown off by them, we chose to focus exclusively on receptor sequence signals in the simulated datasets.
", "rawMarkdown": "Good observation! While the simulation of receptor sequence signals can also lean on insights from studies of antigen binding of individual receptors, there is no similar external basis to lean on regarding the clonal expansion of disease-relevant cells - in particular for the interaction between sequence signals and clonal expansion. To avoid any methods either exploiting unrealistically simulated clonal expansion patterns or getting thrown off by them, we chose to focus exclusively on receptor sequence signals in the simulated datasets.", "votes": null }, { "id": "3312101", "postDate": "11/06/2025 13:05:58", "content": "Thank you, but that seems to be answering another question. As you show in the \"real\" datasets you provide, not all clones are of the same sizes. By ignoring this in simulated data, you ignore a core behavior of the immune system (it does not create new T cells, it regulates their sizes). This would be like simulating an economy with people's identification number and not their income.
", "rawMarkdown": "Thank you, but that seems to be answering another question. As you show in the \"real\" datasets you provide, not all clones are of the same sizes. By ignoring this in simulated data, you ignore a core behavior of the immune system (it does not create new T cells, it regulates their sizes). This would be like simulating an economy with people's identification number and not their income.", "votes": null }, { "id": "3312132", "postDate": "11/06/2025 14:21:46", "content": "I would say it is more like analysing people based on which shops they visit, but without data on how much they spend, if any.
", "rawMarkdown": "I would say it is more like analysing people based on which shops they visit, but without data on how much they spend, if any.", "votes": null }, { "id": "3312514", "postDate": "11/07/2025 11:11:45", "content": "Thanks for the good points! We provide arguments below for why we think focusing solely on discovering label-associated sequence patterns is still an important subtask:
\nCurrently, most methods in the field do not use clonal counts for repertoire classification, despite the fact that the immune system responds through clonal expansion. We believe this choice of ignoring clonal counts is still a reasonable approach because a sequenced repertoire is only a small sample of the underlying immune cell population; any immune cell that is clonally expanded will have a substantially increased chance of being part of a sequenced repertoire - thus, presence in a sequenced repertoire indicates a higher likelihood of clonal expansion in the underlying immune cell population.
When methods in the field have used clonal counts, these counts have had a supporting role in the discovery of label-associated sequence patterns for repertoire classification. Even for this line of methodology, identifying sequence patterns thus constitutes a primary challenge.
This seems correct:
\n\n\nany immune cell that is clonally expanded will have a substantially increased chance of being part of a sequenced repertoire
\n
But the following is not the consequence of the previous part (i.e., A -> B doesn't mean that B -> A):
\n\n\nthus, presence in a sequenced repertoire indicates a higher likelihood of clonal expansion in the underlying immune cell population
\n
because we all know that immune repertoires have an extremely heavy tail of low-abundance receptors. While any significantly expanded cell will be observed in the repertoire, without counts, we can't really infer which cells are more significantly expanded and which are just noise from the heavy tail.
", "rawMarkdown": "This seems correct:\n\n> any immune cell that is clonally expanded will have a substantially increased chance of being part of a sequenced repertoire\n\nBut the following is not the consequence of the previous part (i.e., A -> B doesn't mean that B -> A):\n\n> thus, presence in a sequenced repertoire indicates a higher likelihood of clonal expansion in the underlying immune cell population\n\nbecause we all know that immune repertoires have an extremely heavy tail of low-abundance receptors. While any significantly expanded cell will be observed in the repertoire, without counts, we can't really infer which cells are more significantly expanded and which are just noise from the heavy tail.", "votes": null }, { "id": "3312878", "postDate": "11/08/2025 06:18:21", "content": "Yes, we agree. What we mean is that for a given clonotype, being clonally expanded increases its probability of being included in the sequenced repertoire. So, clonal expansion leaves an imprint in the sequenced repertoires. And we agree that the heavy tail of low-abundance clonotypes will still make up a substantial proportion of the sequenced repertoires.
", "rawMarkdown": "Yes, we agree. What we mean is that for a given clonotype, being clonally expanded increases its probability of being included in the sequenced repertoire. So, clonal expansion leaves an imprint in the sequenced repertoires. And we agree that the heavy tail of low-abundance clonotypes will still make up a substantial proportion of the sequenced repertoires.", "votes": null } ], "comments": [ { "id": 3312058, "author_name": "ckanduri", "author_url": "", "post_date": "11/06/2025 10:47:49", "content": "Good observation! While the simulation of receptor sequence signals can also lean on insights from studies of antigen binding of individual receptors, there is no similar external basis to lean on regarding the clonal expansion of disease-relevant cells - in particular for the interaction between sequence signals and clonal expansion. To avoid any methods either exploiting unrealistically simulated clonal expansion patterns or getting thrown off by them, we chose to focus exclusively on receptor sequence signals in the simulated datasets.
", "votes": null, "replies": [ { "id": 3312101, "author_name": "parejepareje", "author_url": "", "post_date": "11/06/2025 13:05:58", "content": "Thank you, but that seems to be answering another question. As you show in the \"real\" datasets you provide, not all clones are of the same sizes. By ignoring this in simulated data, you ignore a core behavior of the immune system (it does not create new T cells, it regulates their sizes). This would be like simulating an economy with people's identification number and not their income.
", "votes": null, "replies": [ { "id": 3312132, "author_name": "vadimnazarov", "author_url": "", "post_date": "11/06/2025 14:21:46", "content": "I would say it is more like analysing people based on which shops they visit, but without data on how much they spend, if any.
", "votes": null, "replies": [] }, { "id": 3312514, "author_name": "ckanduri", "author_url": "", "post_date": "11/07/2025 11:11:45", "content": "Thanks for the good points! We provide arguments below for why we think focusing solely on discovering label-associated sequence patterns is still an important subtask:
\nCurrently, most methods in the field do not use clonal counts for repertoire classification, despite the fact that the immune system responds through clonal expansion. We believe this choice of ignoring clonal counts is still a reasonable approach because a sequenced repertoire is only a small sample of the underlying immune cell population; any immune cell that is clonally expanded will have a substantially increased chance of being part of a sequenced repertoire - thus, presence in a sequenced repertoire indicates a higher likelihood of clonal expansion in the underlying immune cell population.
When methods in the field have used clonal counts, these counts have had a supporting role in the discovery of label-associated sequence patterns for repertoire classification. Even for this line of methodology, identifying sequence patterns thus constitutes a primary challenge.
This seems correct:
\n\n\nany immune cell that is clonally expanded will have a substantially increased chance of being part of a sequenced repertoire
\n
But the following is not the consequence of the previous part (i.e., A -> B doesn't mean that B -> A):
\n\n\nthus, presence in a sequenced repertoire indicates a higher likelihood of clonal expansion in the underlying immune cell population
\n
because we all know that immune repertoires have an extremely heavy tail of low-abundance receptors. While any significantly expanded cell will be observed in the repertoire, without counts, we can't really infer which cells are more significantly expanded and which are just noise from the heavy tail.
", "votes": null, "replies": [ { "id": 3312878, "author_name": "ckanduri", "author_url": "", "post_date": "11/08/2025 06:18:21", "content": "Yes, we agree. What we mean is that for a given clonotype, being clonally expanded increases its probability of being included in the sequenced repertoire. So, clonal expansion leaves an imprint in the sequenced repertoires. And we agree that the heavy tail of low-abundance clonotypes will still make up a substantial proportion of the sequenced repertoires.
", "votes": null, "replies": [] } ] } ] } ] } ] } ], "raw_markdown_by_id": { "3311695": "What I assume to be simulated data - train_dataset1 - has clone sizes that are very different from real clones. the largest clone apears twice. Why is that? Is it not supposed to be repertoire data?", "3312058": "Good observation! While the simulation of receptor sequence signals can also lean on insights from studies of antigen binding of individual receptors, there is no similar external basis to lean on regarding the clonal expansion of disease-relevant cells - in particular for the interaction between sequence signals and clonal expansion. To avoid any methods either exploiting unrealistically simulated clonal expansion patterns or getting thrown off by them, we chose to focus exclusively on receptor sequence signals in the simulated datasets.", "3312101": "Thank you, but that seems to be answering another question. As you show in the \"real\" datasets you provide, not all clones are of the same sizes. By ignoring this in simulated data, you ignore a core behavior of the immune system (it does not create new T cells, it regulates their sizes). This would be like simulating an economy with people's identification number and not their income.", "3312132": "I would say it is more like analysing people based on which shops they visit, but without data on how much they spend, if any.", "3312514": "Thanks for the good points! We provide arguments below for why we think focusing solely on discovering label-associated sequence patterns is still an important subtask:\n\n- Currently, most methods in the field do not use clonal counts for repertoire classification, despite the fact that the immune system responds through clonal expansion. We believe this choice of ignoring clonal counts is still a reasonable approach because a sequenced repertoire is only a small sample of the underlying immune cell population; any immune cell that is clonally expanded will have a substantially increased chance of being part of a sequenced repertoire - thus, presence in a sequenced repertoire indicates a higher likelihood of clonal expansion in the underlying immune cell population.\n\n- When methods in the field have used clonal counts, these counts have had a supporting role in the discovery of label-associated sequence patterns for repertoire classification. Even for this line of methodology, identifying sequence patterns thus constitutes a primary challenge.", "3312573": "This seems correct:\n\n> any immune cell that is clonally expanded will have a substantially increased chance of being part of a sequenced repertoire\n\nBut the following is not the consequence of the previous part (i.e., A -> B doesn't mean that B -> A):\n\n> thus, presence in a sequenced repertoire indicates a higher likelihood of clonal expansion in the underlying immune cell population\n\nbecause we all know that immune repertoires have an extremely heavy tail of low-abundance receptors. While any significantly expanded cell will be observed in the repertoire, without counts, we can't really infer which cells are more significantly expanded and which are just noise from the heavy tail.", "3312878": "Yes, we agree. What we mean is that for a given clonotype, being clonally expanded increases its probability of being included in the sequenced repertoire. So, clonal expansion leaves an imprint in the sequenced repertoires. And we agree that the heavy tail of low-abundance clonotypes will still make up a substantial proportion of the sequenced repertoires." }, "source": "meta" }