#biodata22 — Public Fediverse posts

Short thread on my poster from #biodata22 on detecting allelic imbalance at isoform-level and in single cells, work with Rob Patro, Noor Singh, Euphy Wu et al.

PDF here: https://www.dropbox.com/s/yjr0d4mndnozwmd/DATA_22_Love.pdf?dl=0

Short thread on my poster from #biodata22 on detecting allelic imbalance at isoform-level and in single cells, work with Rob Patro, Noor Singh, Euphy Wu et al.

PDF here: https://www.dropbox.com/s/yjr0d4mndnozwmd/DATA_22_Love.pdf?dl=0

Short thread on my poster from #biodata22 on detecting allelic imbalance at isoform-level and in single cells, work with Rob Patro, Noor Singh, Euphy Wu et al.

PDF here: https://www.dropbox.com/s/yjr0d4mndnozwmd/DATA_22_Love.pdf?dl=0

Short thread on my poster from #biodata22 on detecting allelic imbalance at isoform-level and in single cells, work with Rob Patro, Noor Singh, Euphy Wu et al.

PDF here: https://www.dropbox.com/s/yjr0d4mndnozwmd/DATA_22_Love.pdf?dl=0

Short thread on my poster from #biodata22 on detecting allelic imbalance at isoform-level and in single cells, work with Rob Patro, Noor Singh, Euphy Wu et al.

PDF here: https://www.dropbox.com/s/yjr0d4mndnozwmd/DATA_22_Love.pdf?dl=0

@timtriche yeah I’ve been manually cross posting.

At #biodata22 I found Twitter was easier to use, eg I want to quickly look up handles and draft posts / threads during sessions, mostly to promote work by Phd students. Both of those are hard to do here (the latter not possible w the main app).

But for me this is week 1 of trying a new thing, entirely OSS and hosted/moderated by volunteers so I’ve got lots of patience to figure things out

@timtriche yeah I’ve been manually cross posting.

At #biodata22 I found Twitter was easier to use, eg I want to quickly look up handles and draft posts / threads during sessions, mostly to promote work by Phd students. Both of those are hard to do here (the latter not possible w the main app).

But for me this is week 1 of trying a new thing, entirely OSS and hosted/moderated by volunteers so I’ve got lots of patience to figure things out

@timtriche yeah I’ve been manually cross posting.

At #biodata22 I found Twitter was easier to use, eg I want to quickly look up handles and draft posts / threads during sessions, mostly to promote work by Phd students. Both of those are hard to do here (the latter not possible w the main app).

But for me this is week 1 of trying a new thing, entirely OSS and hosted/moderated by volunteers so I’ve got lots of patience to figure things out

@timtriche yeah I’ve been manually cross posting.

At #biodata22 I found Twitter was easier to use, eg I want to quickly look up handles and draft posts / threads during sessions, mostly to promote work by Phd students. Both of those are hard to do here (the latter not possible w the main app).

But for me this is week 1 of trying a new thing, entirely OSS and hosted/moderated by volunteers so I’ve got lots of patience to figure things out

That's a wrap on #biodata22, next one is #biodata24 on November 6-9, 2024.

Wish I could've attended in person this year, but a quick shout out to all the organizers who enabled a quick pivot to virtual! You da real MVPs!

That's a wrap on #biodata22, next one is #biodata24 on November 6-9, 2024.

Wish I could've attended in person this year, but a quick shout out to all the organizers who enabled a quick pivot to virtual! You da real MVPs!

That's a wrap on #biodata22, next one is #biodata24 on November 6-9, 2024.

Wish I could've attended in person this year, but a quick shout out to all the organizers who enabled a quick pivot to virtual! You da real MVPs!

Markus Sommer #biodata22 closing us out with "Structure‐guided isoform analysis for the human transcriptome".

Problem: We have many more transcript annotations than genes. Which isoforms actually represent functional proteins?

Leveraging folding algorithms (e.g. AlphaFold2) to score each isoform, high score = more likely to be functional. Showed some examples where this scoring approach matches experimental data. Says not perfect, but helpful data point.

Website: https://www.isoform.io/

Markus Sommer #biodata22 closing us out with "Structure‐guided isoform analysis for the human transcriptome".

Problem: We have many more transcript annotations than genes. Which isoforms actually represent functional proteins?

Leveraging folding algorithms (e.g. AlphaFold2) to score each isoform, high score = more likely to be functional. Showed some examples where this scoring approach matches experimental data. Says not perfect, but helpful data point.

Website: https://www.isoform.io/

Markus Sommer #biodata22 closing us out with "Structure‐guided isoform analysis for the human transcriptome".

Problem: We have many more transcript annotations than genes. Which isoforms actually represent functional proteins?

Leveraging folding algorithms (e.g. AlphaFold2) to score each isoform, high score = more likely to be functional. Showed some examples where this scoring approach matches experimental data. Says not perfect, but helpful data point.

Website: https://www.isoform.io/

Harun Mustafa #biodata22 on "A modular multi‐label framework for aligning sequences to large read set databases and (pan)genomes".

Problem: Low-coverage pan-sample (or genome?) alignment is challenging due to gaps in graphs (both sequence and labels), leading to shorter alignments downstream.

Describing a method, "MetaGraph-MLA", that allows the aligner to leverage "similar" samples (i.e. without the gap) to increase alignment lengths.

Pre-print: https://www.biorxiv.org/content/10.1101/2022.11.04.514718v1

Harun Mustafa #biodata22 on "A modular multi‐label framework for aligning sequences to large read set databases and (pan)genomes".

Problem: Low-coverage pan-sample (or genome?) alignment is challenging due to gaps in graphs (both sequence and labels), leading to shorter alignments downstream.

Describing a method, "MetaGraph-MLA", that allows the aligner to leverage "similar" samples (i.e. without the gap) to increase alignment lengths.

Pre-print: https://www.biorxiv.org/content/10.1101/2022.11.04.514718v1

Katharine Jenike #biodata22 on "Establishing a Solanum pan‐genome to dissect dynamics of paralog evolution".

Building a pan-genome from Solanum, so far with 17 fully assembled genomes built via HiFi + hifiasm followed by scaffolding via HiC and Bionano. Assemblies are chromosome scale and then annotated.

Described "Panagram", a tool for visualizing the constructed pangenome and exploring unique sequence, synteny, etc.:
https://github.com/kjenike/Panagram

Katharine Jenike #biodata22 on "Establishing a Solanum pan‐genome to dissect dynamics of paralog evolution".

Building a pan-genome from Solanum, so far with 17 fully assembled genomes built via HiFi + hifiasm followed by scaffolding via HiC and Bionano. Assemblies are chromosome scale and then annotated.

Described "Panagram", a tool for visualizing the constructed pangenome and exploring unique sequence, synteny, etc.:
https://github.com/kjenike/Panagram

Katharine Jenike #biodata22 on "Establishing a Solanum pan‐genome to dissect dynamics of paralog evolution".

Building a pan-genome from Solanum, so far with 17 fully assembled genomes built via HiFi + hifiasm followed by scaffolding via HiC and Bionano. Assemblies are chromosome scale and then annotated.

Described "Panagram", a tool for visualizing the constructed pangenome and exploring unique sequence, synteny, etc.:
https://github.com/kjenike/Panagram

Katharine Jenike #biodata22 on "Establishing a Solanum pan‐genome to dissect dynamics of paralog evolution".

Building a pan-genome from Solanum, so far with 17 fully assembled genomes built via HiFi + hifiasm followed by scaffolding via HiC and Bionano. Assemblies are chromosome scale and then annotated.

Described "Panagram", a tool for visualizing the constructed pangenome and exploring unique sequence, synteny, etc.:
https://github.com/kjenike/Panagram

Robert Patro (@rob) #biodata22 on "Keeping k‐mers in check—Building fast, small, and composable indices based on the De Bruijn graph".

Problem: Reference indexing is challenging, as we add reference (e.g. pangenome), the index grows rapidly. How do we keep this reference small?

Suggests model that splits index into two "tables" allowing for modular implementations, isolating bottlenecks.

Two repos mentioned:
Piscem: https://github.com/COMBINE-lab/piscem
Pufferfish2: https://github.com/COMBINE-lab/pufferfish2

Robert Patro (@rob) #biodata22 on "Keeping k‐mers in check—Building fast, small, and composable indices based on the De Bruijn graph".

Problem: Reference indexing is challenging, as we add reference (e.g. pangenome), the index grows rapidly. How do we keep this reference small?

Suggests model that splits index into two "tables" allowing for modular implementations, isolating bottlenecks.

Two repos mentioned:
Piscem: https://github.com/COMBINE-lab/piscem
Pufferfish2: https://github.com/COMBINE-lab/pufferfish2

Robert Patro (@rob) #biodata22 on "Keeping k‐mers in check—Building fast, small, and composable indices based on the De Bruijn graph".

Problem: Reference indexing is challenging, as we add reference (e.g. pangenome), the index grows rapidly. How do we keep this reference small?

Suggests model that splits index into two "tables" allowing for modular implementations, isolating bottlenecks.

Two repos mentioned:
Piscem: https://github.com/COMBINE-lab/piscem
Pufferfish2: https://github.com/COMBINE-lab/pufferfish2

Robert Patro (@rob) #biodata22 on "Keeping k‐mers in check—Building fast, small, and composable indices based on the De Bruijn graph".

Problem: Reference indexing is challenging, as we add reference (e.g. pangenome), the index grows rapidly. How do we keep this reference small?

Suggests model that splits index into two "tables" allowing for modular implementations, isolating bottlenecks.

Two repos mentioned:
Piscem: https://github.com/COMBINE-lab/piscem
Pufferfish2: https://github.com/COMBINE-lab/pufferfish2

@rob new implementation Piscem, already in use in different contexts worked on by the lab #biodata22

@rob new implementation Piscem, already in use in different contexts worked on by the lab #biodata22

@rob new implementation Piscem, already in use in different contexts worked on by the lab #biodata22

@rob new implementation Piscem, already in use in different contexts worked on by the lab #biodata22

@rob maps kmer to tile, tile to occurrences #biodata22

@rob maps kmer to tile, tile to occurrences #biodata22

@rob maps kmer to tile, tile to occurrences #biodata22

@rob maps kmer to tile, tile to occurrences #biodata22

@rob allowing toots today at #biodata22

k-mer based reference indexing

@rob allowing toots today at #biodata22

k-mer based reference indexing

@rob allowing toots today at #biodata22

k-mer based reference indexing

@rob allowing toots today at #biodata22

k-mer based reference indexing

Haoyu Cheng #biodata22 on "An integrated algorithm for robust and cost‐effective telomere‐to‐telomere genome assembly".

Describing approach to perform hybrid assembly by combining PB HiFi reads (accurate) with ONT ultra-long reads (noisy, but long). Many challenges discussed

End result seems to be that using HiFi backbone, the UL help extend the assembly into longer N50. A few T2T chromosomes finish automatically, and those that don't have few gaps.

Haoyu Cheng #biodata22 on "An integrated algorithm for robust and cost‐effective telomere‐to‐telomere genome assembly".

Describing approach to perform hybrid assembly by combining PB HiFi reads (accurate) with ONT ultra-long reads (noisy, but long). Many challenges discussed

End result seems to be that using HiFi backbone, the UL help extend the assembly into longer N50. A few T2T chromosomes finish automatically, and those that don't have few gaps.

Talk was also filled with plenty of punny D&D jokes! #biodata22

Talk was also filled with plenty of punny D&D jokes! #biodata22

Jessica Bonnie #biodata22 with "DandD—Utilizing “Delta delta” (Δδ) to quantify novel contributions from genomes".

Problem: How do we measure "new" sequence in an assembly? Current approaches require parameter choices such as aligner or "k"-mer choice.

DandD method uses information theory to measure change in compressibility as you add more assemblies. Benefits are alignment-free, parameter-free, and efficient to compute. Also allows for similarity scoring via modified delta-delta Jaccard.

Jessica Bonnie #biodata22 with "DandD—Utilizing “Delta delta” (Δδ) to quantify novel contributions from genomes".

Problem: How do we measure "new" sequence in an assembly? Current approaches require parameter choices such as aligner or "k"-mer choice.

DandD method uses information theory to measure change in compressibility as you add more assemblies. Benefits are alignment-free, parameter-free, and efficient to compute. Also allows for similarity scoring via modified delta-delta Jaccard.

Tavor Baharav #biodata22 with "A statistical reference‐free genomic algorithm subsumes common workflows and enables novel discovery".

(Audio returned halfway through)

Repo: https://github.com/salzman-lab/nomad

Tavor Baharav #biodata22 with "A statistical reference‐free genomic algorithm subsumes common workflows and enables novel discovery".

(Audio returned halfway through)

Repo: https://github.com/salzman-lab/nomad

Last day of #biodata22 starts with Elinor Karlsson with "Leveraging base pair mammalian constraint to understand mammalian evolution and human disease".

(Virtual attendees basically missed this entire talk due to audio technical issues 😬, BUT I will say it looked like there were lots of great visuals by the presenter!)

Last day of #biodata22 starts with Elinor Karlsson with "Leveraging base pair mammalian constraint to understand mammalian evolution and human disease".

(Virtual attendees basically missed this entire talk due to audio technical issues 😬, BUT I will say it looked like there were lots of great visuals by the presenter!)

It's fun to watch visual presentations without audio and make up your own stories about what the presenter is trying to say. I'm probably about 10% correct 😆 #biodata22

It's fun to watch visual presentations without audio and make up your own stories about what the presenter is trying to say. I'm probably about 10% correct 😆 #biodata22

It's fun to watch visual presentations without audio and make up your own stories about what the presenter is trying to say. I'm probably about 10% correct 😆 #biodata22

It's fun to watch visual presentations without audio and make up your own stories about what the presenter is trying to say. I'm probably about 10% correct 😆 #biodata22

Keynote Molly Przeworski #biodata22 on "Causes and consequences of recombination hotspots in vertebrates".

Presenting on years of work studying recombination hotspots. Focusing on PRDM9 which seems to help generate hotspots and make recombination efficient. Data showing that PRDM9 was adopted before the origin of vertebrates and has been independently loss (entirely/partially) in many species. Idea that PRDM9 has to evolve rapidly to maintain benefit of keeping recombination more efficient.

Keynote Molly Przeworski #biodata22 on "Causes and consequences of recombination hotspots in vertebrates".

Presenting on years of work studying recombination hotspots. Focusing on PRDM9 which seems to help generate hotspots and make recombination efficient. Data showing that PRDM9 was adopted before the origin of vertebrates and has been independently loss (entirely/partially) in many species. Idea that PRDM9 has to evolve rapidly to maintain benefit of keeping recombination more efficient.

Jingyou Rao #biodata22 with "Computational approaches for inferring gene regulation in in situ perturbation screens".

Focusing on using in situ imaging perturbation as opposed to singe-cell sequencing counterpart. This has less experimental dropouts and lower dimensionality. Created sparse and dense models for different biological situations and tested with simulation data to show trade-offs.

Jingyou Rao #biodata22 with "Computational approaches for inferring gene regulation in in situ perturbation screens".

Focusing on using in situ imaging perturbation as opposed to singe-cell sequencing counterpart. This has less experimental dropouts and lower dimensionality. Created sparse and dense models for different biological situations and tested with simulation data to show trade-offs.

Andrea Rendeiro #biodata22 on "Unsupervised discovery of tissue architecture in multiplexed imaging".

(I missed the intro, sorry!)

Describing a method called UTAG (unsupervised discovery of tissue architectures in graphs) that converts images in tissue microanatomical domains via segmentation and graph algorithms.

Paper: https://www.nature.com/articles/s41592-022-01657-2
Repo: https://github.com/ElementoLab/utag

Andrea Rendeiro #biodata22 on "Unsupervised discovery of tissue architecture in multiplexed imaging".

(I missed the intro, sorry!)

Describing a method called UTAG (unsupervised discovery of tissue architectures in graphs) that converts images in tissue microanatomical domains via segmentation and graph algorithms.

Paper: https://www.nature.com/articles/s41592-022-01657-2
Repo: https://github.com/ElementoLab/utag

Atishay Jain #biodata22 with "Scalable and memory efficient segmentation of large microscopy images using graph‐based neural networks".

Goal is to segment an image into cell or not-cell (binary). Problem is that images are large, and current approaches consume too much memory on GPU.

New approach that converts images into a graph (via superpixels) and then fed to a GNN. Problem goes from millions of pixels to thousands of nodes. Slight reduction in accuracy, but big gains in performance.

Atishay Jain #biodata22 with "Scalable and memory efficient segmentation of large microscopy images using graph‐based neural networks".

Goal is to segment an image into cell or not-cell (binary). Problem is that images are large, and current approaches consume too much memory on GPU.

New approach that converts images into a graph (via superpixels) and then fed to a GNN. Problem goes from millions of pixels to thousands of nodes. Slight reduction in accuracy, but big gains in performance.

Atishay Jain #biodata22 with "Scalable and memory efficient segmentation of large microscopy images using graph‐based neural networks".

Goal is to segment an image into cell or not-cell (binary). Problem is that images are large, and current approaches consume too much memory on GPU.

New approach that converts images into a graph (via superpixels) and then fed to a GNN. Problem goes from millions of pixels to thousands of nodes. Slight reduction in accuracy, but big gains in performance.