Michael Lazear @michaellazear.bsky.social · 09/10/2026How have you been solving them?! I've just been playing video games this whole time... 010
Michael Lazear @michaellazear.bsky.social · 09/10/2026Back when I did some of that fun OS stuff, I actually booted it off of a physical floppy drive! 110
Michael Lazear @michaellazear.bsky.social · 09/10/2026I have considered a rewrite so that it can run on a toaster... 120
Michael Lazear @michaellazear.bsky.social · 09/10/2026This is actually what got me into programming ~20 years ago - there is a fantastic website (wiki.osdev.org/Expanded_Mai...) for hobby operating systems. IMO one of the best projects (along with compilers) for developing a mastery of programming and how computers workosdev.orgOS Development 120
Michael Lazear @michaellazear.bsky.social · 07/10/2026Are you willing to share the full training/test/val dataset? 100
Michael Lazear @michaellazear.bsky.social · 07/10/2026I saw the results in the paper, but if you spent 4,800 additional GPU hours on training Casanovo end-to-end, does it beat dIon initialization? Without compute-matched baselines it's harder to interpret how much the self-supervision is really helping 100
Michael Lazear @michaellazear.bsky.social · 07/10/2026Very cool work! How does this compare to just training Casanovo from scratch at an equivalent parameter or FLOPS budget? 100
Michael Lazear @michaellazear.bsky.social · 07/10/2026At ~4.35M vCPU hours logged, that is a LOT of $$$ in compute saved for users versus the next-fastest tools! 030
Michael Lazear @michaellazear.bsky.social · 07/10/2026As part of my overhaul of Sage's automated testing infrastructure I put together a dashboard that aggregates the anonymized telemetry reports Sage collections - great timing, because Sage just hit 1,000,000 logged searches globally. sage.lazear.org/usage 121
Michael Lazear @michaellazear.bsky.social · 05/10/2026Sage already runs on Windows/Linux/Mac, now you can even run it on your phone 060
Michael Lazear @michaellazear.bsky.social · 04/10/2026Have you ever wanted to run a proteomics search engine directly in your browser? No, I haven't either... But now you can! All of your data stays on your local computer, and you can run Sage (with some experimental features 😀) through a nice UI without installing anything. sage.lazear.org/search/ 1152
Michael Lazear @michaellazear.bsky.social · 03/10/2026I think the whole "no more file conversion" thing has already been solved by agents 😄 010
Michael Lazear @michaellazear.bsky.social · 02/10/2026I mean if it's $5, then yeah? just use a personal account? 000
Michael Lazear @michaellazear.bsky.social · 02/10/2026Is this the first Rust -> C# rewrite in history? 110
Michael Lazear @michaellazear.bsky.social · 02/10/2026Charge some nominal fee for a submission - eliminates a lot of very low quality stuff, and also recoups some of the cost for increased traffic/volume 101
Michael Lazear @michaellazear.bsky.social · 30/09/2026Yeah, the major benefit of open source (especially with coding agents) is that everyone can read/understand/modify it for their specific needs! We don't need to wait for the anointed high priests to share their version of the truth with us anymore. 130
Michael Lazear @michaellazear.bsky.social · 30/09/2026I don't think most of those Walmart greeters are doing it as a hobby though... or at least the ones doing it as a hobby probably aren't worried about being made obsolete! I just do this for the love of the game. 120
Michael Lazear @michaellazear.bsky.social · 30/09/2026I'm not too worried about being made irrelevant yet :) 220
Michael Lazear @michaellazear.bsky.social · 30/09/2026what do we need to make Sage (or a next-gen DIA tool) even more agent friendly? MCP access to search engine internals? better docs? 130
Michael Lazear @michaellazear.bsky.social · 26/09/2026Well that's definitely a different scenario than "20% more" - but the point still stands, you probably don't actually need to do anything special. I would just do median polish/VSN/whatever you normally do... 000
Michael Lazear @michaellazear.bsky.social · 26/09/2026If you're just overexpressing a protein, why would you change normalization at all from what you're usually doing? 100
Michael Lazear @michaellazear.bsky.social · 24/09/2026Word on the street is that Sage is the official proteomics search engine of .gov! github.com/usnistgov/sa... 151
Michael Lazear @michaellazear.bsky.social · 22/09/2026The ability to rapidly go from idea -> implementation -> results/learnings is also an insanely enabling teaching tool, if you are so inclined. 020
Michael Lazear @michaellazear.bsky.social · 22/09/2026I agree in large part. LLMs are very powerful tools - as with anything powerful, they require responsible use. While they can definitely be used to shortcut learning, the ability to have interactive tutorials made for basically any subject (with high fidelity) on-the-fly is pretty damn cool 220
Michael Lazear @michaellazear.bsky.social · 08/09/2026RT-if you've never used the internet, period. 000
Michael Lazear @michaellazear.bsky.social · 23/08/2026I like 0->0.9. My computer is littered with projects all stalled at 0.9 141
Reposted by Michael LazearBen Lehner @benlehner.bsky.social · 03/08/20261st genome sequenced (Sanger 1977), 1st genome synthesised (Venter 2003), 1st genome + proteome fully mutated (Huijin Xiangua 2026!) @crg.eu @sangerinstitute.bsky.social Complete Mutagenesis of the Genome and Proteome of ΦX174 www.biorxiv.org/content/10.6...biorxiv.orgComplete Mutagenesis of the Genome and Proteome of ΦX174The bacteriophage ΦX174 was the first genome to be sequenced and the first to be chemically synthesised. Here we present a complete map of the consequences of changing every nucleotide in the ΦX174 ge... 19742
Reposted by Michael LazearWill Fondrie @willfondrie.com · 30/07/2026New @talusbio.bsky.social #AI preprint just dropped 🚀 Meet Ptarmigan-1. It predicts which small molecules bind a protein, and which residues they hit, from sequence and 2D chemistry alone. No structures, no docking, no poses. www.biorxiv.org/content/10.6... Time for a 🧵 (1/13)biorxiv.orgStructure-free, site-resolved contrastive learning extends small-molecule discovery beyond the reach of structure-based modelingVirtual screening asks which molecules, among an enormous space of drug-like chemistry, are worth synthesizing and testing against a protein target. Most modern methods answer by building and scoring ... 2218
Michael Lazear @michaellazear.bsky.social · 27/07/2026The final stage has been so much fun the last two years, couldn't have asked for a more entertaining finish yesterday! 110
Michael Lazear @michaellazear.bsky.social · 17/07/2026it's fast because it if took more than 3 minutes I would get bored, wander off, and forget what I was doing 😀 010
Michael Lazear @michaellazear.bsky.social · 15/07/2026Do you want to work at the frontier of structural and chemical #proteomics? Interested in induced proximity and building a feedback loop for machine learning models to explore biology? I'm hiring! Apply below ⬇️ jobs.ashbyhq.com/proxima/6e39...jobs.ashbyhq.comProteomics ScientistProteomics Scientist 1113
Michael Lazear @michaellazear.bsky.social · 09/07/2026I'm pretty sure that profile/chat history is used for the safeguard classifier and it is super frustrating. Fable doesn't even exist as far as I'm concerned, just an elaborate marketing scheme 010
Michael Lazear @michaellazear.bsky.social · 03/07/2026why do you say you need to remove "common" mods to do quant? I have successfully done (legit) quantification on open-search results many times 120
Michael Lazear @michaellazear.bsky.social · 01/07/2026Manipulation of raw data also seems, to me, much less likely than misuse (unintentional or otherwise) of some other part of the data analysis pipeline - especially when 99% of the field is relying on black box software whose authors refuse to describe their work in detail. 100
Michael Lazear @michaellazear.bsky.social · 01/07/2026I just don't see this being a huge issue in practice, any more than being worried about people spiking in proteins for their WBs (again just comes down to trust). How many scientists are realistically even looking hard at their own raw spectra, let alone other peoples? 100
Michael Lazear @michaellazear.bsky.social · 01/07/2026If we're at the point where I have to cryptographically verify someone's data I would just rather not use their data at all 110
Michael Lazear @michaellazear.bsky.social · 01/07/2026Doesn't this just always come down to trust? Someone really dedicating to faking an experiment would just synthesize their protein/peptides of interest and spike them in a sample. No digital chain of trust can prevent that - but I do feel that the best thing to do is just onboard vendor key signing. 320
Michael Lazear @michaellazear.bsky.social · 10/06/2026Wait until you see how fast my DIA tool is :) 3100
Michael Lazear @michaellazear.bsky.social · 30/05/2026If you're in town for #ASMS2026 and want to talk chemoproteomics, computational proteomics, or generating high-throughput data for training ML models - hit me up! 1101
Reposted by Michael LazearWill Fondrie @willfondrie.com · 30/05/2026Looking for a burrito at #ASMS2026? Definitely hit up The Taco Stand: maps.app.goo.gl/fSYTwvyUCs57...maps.app.goo.gl 242
