👀

How much data is lost in #science? 🚨 New study: 50 - 90% of electron microscopy images never make it into publications. This hidden "lost data" pool can fuel #AI, #education & new discoveries. https://t.co/58Iz4n5g8F @Hitachi_EM #Microscopy #Artificialintelligence #nano

CATNIP for the win! Read our newest work with the Gomes group- https://t.co/lILWdTxbVR @Apatoneh @gabepgomes @daniil_boiko @AlisonNarayanUM

👀

Every person who is working on ML for drug discovery must set up a biological assay themselves. After that you will either reconsider your life choices or understand why the models are not working.

🧲Magnetic stirrers - used daily in millions of labs - may silently sabotage reproducibility. Our @JACS_Au article describes drastic variability in reaction outcomes based on vessel placement https://t.co/cFzO08CBQL @ACSPublications @ACSCatalysis @ChemistryNews @NatureChemistry

you should legally be required to disclose what quantization level you are serving your current model at like it was a nutrition label. you should also be banned from dynamically adjusting quantization based on demand without notification. (you know who you are ...)

Discovering organic reactions with a machine-learning-powered deciphering of tera-scale mass spectrometry data #machinelearning #compchem

Discovering new reactions from terabyte-scale mass spectrometry data with a machine learning approach Large experimental datasets have become a staple of modern chemistry research, yet the sheer volume of stored information often outstrips researchers’ ability to interpret and

How many reasons do you need to move to San Francisco

I guess not citing prior work is all you need…

Most useful website in the internet

(7/n) And finally, thanks 🙌 to all the coauthors and everyone who has supported this work.

(6/n) But this is only beginning and there is still a lot of work to be done. You can play with this platform at https://t.co/EkbxDsiFZR and reach out if you have any questions or excited about this research direction.

(5/n) Having all that in mind, we have built a system that takes a substrate, selects its neighbors in chemical space, looks up known reactivity, collects enzymes and their neighbors in enzyme space, and finally a separate model reranks this list to produce final output. 🤖

(4/n) ... of course, it's not always the case, so some screening will be required. What we want here is that a researcher finds working biocatalyst sooner ⏲️, than later.

(3/n) So, what's next? We need to somehow recommend 🧠enzymes that might work for a particular substrate. Here we exploit two very simple ideas — similar substrates should have similar reactivity and similar enzymes should have similar reactivity as well. And ...

(2/n) For building such a system, one would need a big enough dataset 💾 first. The preprint introduces an enzyme library 📚, aKGLib1, and a dataset, BioCatSet1, that contains information about reactivity of enzymes from the library against a set of more than 100 substrates.

At @gpggrp we published a preprint in collaboration with @NarayanLab and particularly awesome @Apatoneh. #ML #AI #chemistry #science The paper introduces a ML-driven platform for guiding biocatalyst selection in small molecule synthesis. But how does it work? 🧵 (1/n)