Jett Crowdis

🚨🔊Interested in structural variants? Genome sequencing? Pediatric cancers? Germline variants+cancer risk? Rare variants+rare diseases? If so (or even if not!), I'm delighted to share the first paper from my postdoc @VanAllenLab @DanaFarber: biorxiv.org/content/10.110… 1/12

Such a joy to work on this project, and I can't wait to see how it grows!

This was a ton of fun to work on! It pushed me to think outside traditional cancer genomics, and the team is full of caring individuals who genuinely care about sustainably partnering with patients.

@ahmattox @dev_seve Neat stuff! I assume since this tool was built for single-cell, you'd want the embedding + "cell" metadata + matrix values? There are some unique caveats to this data (e.g. the matrix is binary) and we'd need permission to share said matrix, but we could make it happen

@dev_seve @jettcrowdis This looks extremely interesting! Which data in particular were you interested in? We're not able to share the full original data set which you could request from edhrec.com. We also have the map data which is mostly a large CSV of coordinates which I can link.

It's been super fun to work on this in my spare time! Many striking parallels between this project and single-cell genomics (there are about as many Magic cards as genes!). Map: luckypaper.co/resources/comm… Code: github.com/jett-crowdis/m…

@ahmattox @NegativeSilence The algorithm will never allow "isolated" decklists to separate out. There needs to be on the order of ~25 similar decklists for that to happen! It even warns you if a data point is so isolated that it can't find neighbors (as that would throw off other calculations)

@NegativeSilence I think we need @jettcrowdis for a real answer, but the algorithm tends not to leave things isolated. The algorithm isn't geared to preserve negative space and things isolated in 20k space fall in with whatever they're closest to.

I'm excited to finally be publishing the Commander Map! This is a picture of 1.2 million commander decks!! Algorithmically organized by the cards they contain. And you can zoom in and see every last one! luckypaper.co/resources/comm…

@medwar99 @ahmattox Love to see this! Magic + data science + teaching :) I made some choices that are suspect in a research setting (e.g. run clustering on UMAP reduced data), so happy to discuss.

@ahmattox Nice! Do you have a breakdown of your methodology? I have a couple of PhD students working in deep clustering areas and I pay Magic with a different group of students, this will get the two groups talking 😁

For #ProstateCancerAwarenessMonth, an update on the Metastatic Prostate Cancer Project @PrCaProject & bringing patient-driven cancer research to everyone's doorsteps @broadinstitute @DanaFarber @count_me_in @SU2C et al. - please share! mpcproject.org [1/9]

Here's our latest, where we explored the interface of biology and machine learning for cancer genomics discovery in prostate cancer - congrats @HMarakeby et al! @DanaFarber @broadinstitute nature.com/articles/s4158… [1/8]

Super excited to be a part of this - huge shoutout to both @VanAllenLab and @MXresearch for the incredible leadership and figure-making/analytical skills!

1/ Just posted on @biorxivpreprint! Very excited to be sharing our work on Sarcomatoid and Rhabdoid Renal Cell Carcinoma. Incredibly thankful to have been co-mentored by @VanAllenLab & @DrChoueiri! @DanaFarber @broadinstitute @DanaFarberNews biorxiv.org/content/10.110…

@MXresearch @YosefMaruvka @VanAllenLab @nicolasstransky Yep, that would also work!

Here's our latest, led by @jettcrowdis (comp bio extraordinaire) with h/t to old office mate @nicolasstransky & his OG CoMut "nicogram" ca. 2011 (also when he taught me where the 'Applications' folder was on a Mac) See Jett's tweets for info + github + preprint links 👇

@StoverLab @VanAllenLab Thanks Dan - you were definitely part of my inspiration to jump to dry lab, hope you're doing well!

@jettcrowdis Congrats Jett! So awesome to see you continuing to rock, now in the awesome @VanAllenLab

Excited to announce the release of CoMut, a package for creating comutation plots! CoMut is object-oriented, works with arbitrary data types rather than specific genomic data, and has a straightforward interface for adding and plotting data github.com/vanallenlab/co… (1/2)

@YosefMaruvka @VanAllenLab @nicolasstransky Currently it needs to be manually transformed, but please submit a feature request on GitHub if you feel that would be useful!

@jettcrowdis @VanAllenLab @nicolasstransky I meant only on the presentation as a feature of the tool. Without the need to manually transform the input. Thanks

@YosefMaruvka @VanAllenLab @nicolasstransky Definitely - you have complete control over the data added, so it can be in whatever scale you want

@VanAllenLab @jettcrowdis @nicolasstransky Can you enable the number of mutations to be presented on a logarithmic scale?

Check out a notebook that lets you make comut plots without installing anything: colab.research.google.com/github/vanalle… and our paper on bioRxiv biorxiv.org/content/10.110…. Huge thanks to @VanAllenLab for all the support and mentorship! (2/2)