Aaron Jesse O'Hare-Lewis 🏳️‍🌈

As it turns out, the secretion channel Sec61αβγ has a fourth subunit, RAMP4—long overlooked because it falls off in detergent. It holds the pore open and completes the lumenal funnel. This plus exciting structures of TRAP, uL22, & more are now online: biorxiv.org/content/10.110…

Do read the @elife assessment (and the paper too!) of this new story by @aaron_jesse_ et al. "Structural analysis of the dynamic ribosome-translocon complex" doi.org/10.7554/eLife.… Now, tell me you prefer to be judged by where you publish rather than by the quality of your work

We've expanded our RAMP4/TRAP/uL22/etc study with new data confirming biochemically that RAMP4 is a part of the secretory translocon (thanks Frank Zhong & @keenan_lab!). Read it at eLife with an insightful commentary from @FriedrichForst1 & D Vismpas: #x7d9359a9" target="_blank" rel="nofollow noopener">elifesciences.org/articles/98548…

@Haoxi_xiaozzxi Thanks Haoxi! Happy Christmas x

@aaron_jesse_ A series of careful and in-depth analyses! Very insightful and impressive. 🔥

As it turns out, the secretion channel Sec61αβγ has a fourth subunit, RAMP4—long overlooked because it falls off in detergent. It holds the pore open and completes the lumenal funnel. This plus exciting structures of TRAP, uL22, & more are now online: biorxiv.org/content/10.110…

@BJ_Greber (tried to just edit it, so as not to break your link, but apparently that's a paid feature now...)

@BJ_Greber Reposted with the right link, thanks for flagging!

NEW #LMBpublications EMC rectifies the topology of multipass membrane proteins ift.tt/WqNr4QA

The missing piece in the bacterial #ESCRT-III story.. #Vipp1 forms dynamic spiral filaments on membrane! Spirals are springs that drive 3D ring formation in the spiral centre! 😎😵‍💫 Wonderful collaboration with @Colom_D @RouxLab… 1/2 biorxiv.org/content/10.110…

Language Modeling Is Compression paper page: huggingface.co/papers/2309.10… It has long been established that predictive models can be transformed into lossless compressors and vice versa. Incidentally, in recent years, the machine learning community has focused on training increasingly large and powerful self-supervised (language) models. Since these large language models exhibit impressive predictive capabilities, they are well-positioned to be strong compressors. In this work, we advocate for viewing the prediction problem through the lens of compression and evaluate the compression capabilities of large (foundation) models. We show that large language models are powerful general-purpose predictors and that the compression viewpoint provides novel insights into scaling laws, tokenization, and in-context learning. For example, Chinchilla 70B, while trained primarily on text, compresses ImageNet patches to 43.4% and LibriSpeech samples to 16.4% of their raw size, beating domain-specific compressors like PNG (58.5%) or FLAC (30.3%), respectively. Finally, we show that the prediction-compression equivalence allows us to use any compressor (like gzip) to build a conditional generative model.

Remember when `gzip` and kNN bested an LLM at classifying text in a previously unseen language? It turned out the defeated “large language model” was a BERT with ~100M params. Chinchilla-70B can beat PNG at compressing photos. And FLAC at compressing audio.

Let’s start the week in style: just out in Biorxiv one of the most important results from my lab, with the Reinisch lab @YaleMed: “Lipid scrambling is a general feature of protein insertases”. The title pretty much sums it up. Short 🧵(1/13). biorxiv.org/content/10.110…

Our paper is out where we show that extensive angular sampling in template matching (TM) makes particle detection in tomograms much more sensitive. A GPU implementation speeds up the angular search. Short thread about our findings! mdpi.com/1422-0067/24/1… #teamtomo #CryoEM 1/6

Some simple EM processing tasks can be overly complicated, so I wrote starparser to help me out. Sharing it here in case others find it useful. For example, you can remove particles with specific orientations, among 40+ other command-line options. github.com/sami-chaaban/s…

RNA covalently links to proteins: today in @Nature! #RNAylation as new post-translational modification (or #Proteinylation as new post-transcriptional RNA modification!) by @HoeferLab, @JaeschkeLab, @MaikSchauerte. Important during #phage infection! nature.com/articles/s4158…

We usually think of mRNAs carrying the information needed to make a protein (the central dogma...). But how can the encoded protein control the fate of its encoding mRNA? 🤔 Check out our review on the topic 👇 cell.com/molecular-cell…

How did I miss this? "A conserved ribosomal protein has entirely dissimilar structures in different organisms" from september of last year biorxiv.org/content/10.110…

No-one has ever been able to replicate Gregor Mendel's observations of pea plants. They're a little "too perfect", lacking even random statistical noise that would have been expected from small sample sizes. Was it scientific fraud?

Check out our cryoEM result of mRNA editosome in Trypanosoma brucei! Awesome collaboration with Aphasizhev lab from Boston U. Many congrats to Bruce @ShihengLiu⁩ and other coauthors for another wonderful piece of work! science.org/doi/10.1126/sc…

I’m delighted to share our work @alanbrownhms on the structures of axoneme, one of the largest macromolecular machine in nature! This is really a dream coming true! Now we have an atomic view of the core of cilia skeleton. nature.com/articles/s4158…