Shaun Mahony
@shaunmahony.bsky.social
Studying gene regulation and transcription factor binding with machine learning. Assoc Prof at Penn State. 🇮🇪
November 4, 2025 at 12:29 PM
We also see varied associations between ApiAP2 transcription factor binding sites and chromatin states, suggesting a new way to categorize the activities of these TFs.
September 7, 2025 at 3:10 PM
We also see varied associations between ApiAP2 transcription factor binding sites and chromatin states, suggesting a new way to categorize the activities of these TFs.
Chromatin states are highly dynamic during Plasmodium blood stage development; almost two thirds of the genome changes chromatin state during the IDC.
September 7, 2025 at 3:10 PM
Chromatin states are highly dynamic during Plasmodium blood stage development; almost two thirds of the genome changes chromatin state during the IDC.
We found 11 chromatin states during the P. falciparum IDC, defined by combinations of chromatin accessibility and histone modifications. Some of these states comprise unusual combinations of histone marks and associations with regulatory activities that are not seen in higher eukaryotes.
September 7, 2025 at 3:10 PM
We found 11 chromatin states during the P. falciparum IDC, defined by combinations of chromatin accessibility and histone modifications. Some of these states comprise unusual combinations of histone marks and associations with regulatory activities that are not seen in higher eukaryotes.
New preprint from Alan Brown in our lab, who is co-advised by @llinaslab.bsky.social. In this study, Alan studied chromatin state dynamics during the asexual intraerythrocytic development cycle (IDC) of the human malaria parasite, Plasmodium falciparum.
www.biorxiv.org/content/10.1...
www.biorxiv.org/content/10.1...
September 7, 2025 at 3:10 PM
New preprint from Alan Brown in our lab, who is co-advised by @llinaslab.bsky.social. In this study, Alan studied chromatin state dynamics during the asexual intraerythrocytic development cycle (IDC) of the human malaria parasite, Plasmodium falciparum.
www.biorxiv.org/content/10.1...
www.biorxiv.org/content/10.1...
One example: we found a range of sonication settings that provide similar results (at least for CTCF in K562). I kinda expected to see different cohorts of sites in differentially accessible chromatin being "released" by different sonication settings, but we see no evidence for that.
August 23, 2025 at 7:16 PM
One example: we found a range of sonication settings that provide similar results (at least for CTCF in K562). I kinda expected to see different cohorts of sites in differentially accessible chromatin being "released" by different sonication settings, but we see no evidence for that.
MO-ChIP-exo offers improved signal-to-noise compared with other versions of the protocol (as measured by Fraction of Reads in Peaks - FRiP scores).
August 23, 2025 at 7:16 PM
MO-ChIP-exo offers improved signal-to-noise compared with other versions of the protocol (as measured by Fraction of Reads in Peaks - FRiP scores).
Daniela systematically investigated most steps in the protocol. Boxes with green highlights here are steps that were investigated and modified in our new mammalian-optimized ChIP-exo protocol (MO-ChIP-exo).
August 23, 2025 at 7:16 PM
Daniela systematically investigated most steps in the protocol. Boxes with green highlights here are steps that were investigated and modified in our new mammalian-optimized ChIP-exo protocol (MO-ChIP-exo).
Hey @albertvilella.bsky.social - it looks like all your posts are being automatically labeled as spam in my field. Maybe because of the hashtags?
March 5, 2025 at 10:58 PM
Hey @albertvilella.bsky.social - it looks like all your posts are being automatically labeled as spam in my field. Maybe because of the hashtags?
This is a bad take, and demonstrably wrong. Here's the breakdown of F&A for Penn State. Research overheads are more than is recouped from F&A, because of the existing cap on Admin costs.
February 8, 2025 at 3:37 AM
This is a bad take, and demonstrably wrong. Here's the breakdown of F&A for Penn State. Research overheads are more than is recouped from F&A, because of the existing cap on Admin costs.
It's the highlighted condition below I was wondering about. Sounds like journals will need to either charge APCs uniformly or not. NIH-funded authors will not be able to pay OA/APC fees that are only charged to them.
December 24, 2024 at 8:57 PM
It's the highlighted condition below I was wondering about. Sounds like journals will need to either charge APCs uniformly or not. NIH-funded authors will not be able to pay OA/APC fees that are only charged to them.
Well... let's be fair. The new paper doesn't just define the known motifs, it puts them in the context of an integrated predictive model. And the 2013 paper, while great, was far from the first paper to discover this cohort of motifs in promoters. e.g. Xie, et al Nature 2005
May 3, 2024 at 12:56 AM
Well... let's be fair. The new paper doesn't just define the known motifs, it puts them in the context of an integrated predictive model. And the 2013 paper, while great, was far from the first paper to discover this cohort of motifs in promoters. e.g. Xie, et al Nature 2005
Finally, just as they have different affinities for chromatin states, the Fox TFs also have different impacts on the accessibility landscapes. FoxA1, FoxL2, and FoxC1 can open chromatin at their binding sites, while FoxP3 and FoxG1 create no additional accessibility signal. [12/n]
October 28, 2023 at 11:54 PM
Finally, just as they have different affinities for chromatin states, the Fox TFs also have different impacts on the accessibility landscapes. FoxA1, FoxL2, and FoxC1 can open chromatin at their binding sites, while FoxP3 and FoxG1 create no additional accessibility signal. [12/n]
We delve into these attributions at a few individual sites in the paper. I think NN feature attribution is a very useful way to get insight into binding predictors (sequence and chromatin) at individual sites. 11/n
October 28, 2023 at 11:54 PM
We delve into these attributions at a few individual sites in the paper. I think NN feature attribution is a very useful way to get insight into binding predictors (sequence and chromatin) at individual sites. 11/n
But chromatin feature attribs are more informative. For example, the FoxP3 and FoxG1 CNNs have learned that the *lack* of mES ATAC-seq signal is a negative predictor of binding. So the CNNs combine seq and chromatin features to predict binding. 10/n
October 28, 2023 at 11:53 PM
But chromatin feature attribs are more informative. For example, the FoxP3 and FoxG1 CNNs have learned that the *lack* of mES ATAC-seq signal is a negative predictor of binding. So the CNNs combine seq and chromatin features to predict binding. 10/n
We then used feature attribution approaches to ask what features the CNNs use to predict Fox binding. It's clear that seq features don't do the job alone - e.g., the FoxG1 CNN thinks that all Fox sites should be bound based on seq features alone. 9/n
October 28, 2023 at 11:53 PM
We then used feature attribution approaches to ask what features the CNNs use to predict Fox binding. It's clear that seq features don't do the job alone - e.g., the FoxG1 CNN thinks that all Fox sites should be bound based on seq features alone. 9/n
While still not perfect, seq+chromatin improves the representation of Fox binding sites, especially for FoxP3 and FoxG1 (the ones that prefer to bind "active" sites). 8/n
October 28, 2023 at 11:45 PM
While still not perfect, seq+chromatin improves the representation of Fox binding sites, especially for FoxP3 and FoxG1 (the ones that prefer to bind "active" sites). 8/n
To understand how seq and chromatin preferences together specify binding patterns, we turned to CNNs. We trained a convolutional tower architecture to predict the ChIP-exo read counts, using either seq features alone or seq + chromatin features. 7/n
October 28, 2023 at 11:44 PM
To understand how seq and chromatin preferences together specify binding patterns, we turned to CNNs. We trained a convolutional tower architecture to predict the ChIP-exo read counts, using either seq features alone or seq + chromatin features. 7/n
FoxP3 and FoxG1 prefer to bind to sites that were already open and "active" in mES cells, while FoxL2 and FoxC1 have a higher preference for "quiescent" chromatin. The pioneer FoxA1 is in the middle. 6/n
October 28, 2023 at 11:43 PM
FoxP3 and FoxG1 prefer to bind to sites that were already open and "active" in mES cells, while FoxL2 and FoxC1 have a higher preference for "quiescent" chromatin. The pioneer FoxA1 is in the middle. 6/n
If seq features can't fully explain, what about chromatin? Remember, we expressed all Fox TFs in mES cells, so we can compare the induced TFs' binding sites to the pre-existing chromatin features from mES. This is where we find a big difference. 5/n
October 28, 2023 at 11:42 PM
If seq features can't fully explain, what about chromatin? Remember, we expressed all Fox TFs in mES cells, so we can compare the induced TFs' binding sites to the pre-existing chromatin features from mES. This is where we find a big difference. 5/n
While the Fox TFs have similar in vitro motifs, they bind to very different sites on the genome. Sequence features alone do not explain the binding differences (MEME motifs shown, but we tried discriminative models too). 4/n
October 28, 2023 at 11:41 PM
While the Fox TFs have similar in vitro motifs, they bind to very different sites on the genome. Sequence features alone do not explain the binding differences (MEME motifs shown, but we tried discriminative models too). 4/n
Other Fox TFs are sometimes assumed to be pioneers based on their shared DBD (which is similar to the linker histone), but it's not clear if they are. To investigate, we expressed a selection of Fox TFs in the shared chromatin environment of mES cells and measured binding with ChIP-exo. 3/n
October 28, 2023 at 11:40 PM
Other Fox TFs are sometimes assumed to be pioneers based on their shared DBD (which is similar to the linker histone), but it's not clear if they are. To investigate, we expressed a selection of Fox TFs in the shared chromatin environment of mES cells and measured binding with ChIP-exo. 3/n
Looking forward to reading more in depth. First critique comes from my 9yo, who thinks your UMAP looks like a pigeon hocking a huge glob of spit while sitting on someone's head.
October 6, 2023 at 12:56 AM
Looking forward to reading more in depth. First critique comes from my 9yo, who thinks your UMAP looks like a pigeon hocking a huge glob of spit while sitting on someone's head.
Now, where I think we can both agree is, given the reviews and responses you highlighted, this should absolutely not be the editorial summary:
If eLife want to make the new model work, they should give better summaries of the review outcome.
If eLife want to make the new model work, they should give better summaries of the review outcome.
September 26, 2023 at 9:35 PM
Now, where I think we can both agree is, given the reviews and responses you highlighted, this should absolutely not be the editorial summary:
If eLife want to make the new model work, they should give better summaries of the review outcome.
If eLife want to make the new model work, they should give better summaries of the review outcome.
Noticed something when debugging a feed. By default, bluesky is filtering your feeds by language. If someone reposts/replies with just emojis (e.g., to tag something for a feed), it may be filtered because the language isn't detected.
To turn the filters off, just select no language in settings:
To turn the filters off, just select no language in settings:
September 24, 2023 at 8:40 PM
Noticed something when debugging a feed. By default, bluesky is filtering your feeds by language. If someone reposts/replies with just emojis (e.g., to tag something for a feed), it may be filtered because the language isn't detected.
To turn the filters off, just select no language in settings:
To turn the filters off, just select no language in settings: