Srijit Seal 🇺🇸🇬🇧🇩🇪🇯🇵🇮🇳
Senior Chemøinformatician at Merck | Postdoc@Broad Institute of MIT and Harvard, Cambridge(MA) | PhD@University of Cambridge(UK) | AI, Image Analysis, bioML, -omics data, and Cell Painting for drug discovery. srijitseal.com/tools
- So, you want to replace animal models with AI? We spent the last two years collaboratively exploring AI for toxicology with 33 scientists from 14 leading pharmaceutical companies and 6 research institutes! 🏆 American Chemical Society Editor's Choice! pubs.acs.org/doi/10.1021/...
-
View full threadWe spoke to government experts from the National Institute of Environmental Health Sciences (NIEHS) and thought leaders from academia, including the University of Cambridge, the Broad Institute of MIT, and Harvard. This unprecedented industry-wide effort outlines a unified framework for adopting AI.
- This initiative aligns with the FDA's recent announcement to phase out traditional animal testing in favor of more human-relevant methods. Our collective work provides a roadmap for integrating AI, ensuring safer and more efficient drug development processes.
- It’s a strategic blueprint, introducing pillars for success in the real world, drawing from the expertise of scientists from pharma, including GSK, Novartis, Eli Lilly and Company, Merck, AstraZeneca, Bayer | Crop Science, Pfizer, Recursion, Novo Nordisk, Sanofi, Relay Therapeutics, and others.
- A model evaluated 234 toxic and 596 non-toxic compounds using structural fingerprints and achieved an F1 score of 0.440; what could be wrong? Of course, one answer is that it’s a great model. What’s another possibility?
- [Not loaded yet]
- Yesss 🫢
- Today I learned that Hedwig actually died in Harry Potter? How did I even miss this? 😭😭
- [Not loaded yet]
- It so sad! Please tell me the cat survived?
- Thank you for highlighting our work! Use DILIpred to figure out if your compounds have any DILI risk, with mechanistic features :D DILIPred was bulit with over 15k compounds and covers a large chemcial space!
- Drug-induced liver injury (DILI) is a significant concern in drug discovery, in particular because it might only be observed late in large clinical studies. Dilipred is an interesting in-silico tool using variety of approaches to identify potential for hepatotoxicity. macinchem.org/2025/03/02/u...
- Excited to join Merck US as a Senior Research Scientist! If you are in Philly area, would be happy to catchup!
- pip install infoalign! Add biological information to your chemcial fingerprints, using only SMILES as input! I'm thrilled to share that our paper "Learning Molecular Representation in a Cell" has been accepted to ICLR 2025! 🎉
-
View full threadBy combining information bottleneck methods with context graphs, we’re able to extract minimal yet sufficient representations of molecules that lead to better predictions and generalization in downstream tasks like molecular property prediction and zero-shot molecule-morphology matching.
- Many thanks to all authors — John Arevalo, Zhenwen Liang, Anne Carpenter, Meng Jiang, and Shantanu Singh! We look forward to continuing to advance the field of molecular representation learning with. -omics datasets! Find out more! github.com/liugangcode/...
- In this work, Gang Liu introduces InfoAlign, a new approach for learning molecular representations from cellular response data, integrating features like cell morphology and gene expression.
- I did the experiment myself! Having an Android phone saves you 25% costs and the phone pays for itself in 10 Uber rides! Uber and other apps cost up to 30% more on iPhones than on Androids for the same orders. Don’t get me wrong, by all means buy an iPhone! Keep an Android on the side for orders 🤪
- An Android phone offers you more discounts than any subscription model 😂
- It’s not that large is it… hmm
- [Not loaded yet]
- It doesn’t happen to me
- Postdoctoral Opportunity!
- 🎉 I'm starting my own lab at EMBL-EBI (Cambridge, UK; June 2025) 🎉 We will focus on identifying and characterizing chemical hazards to humans and ecosystems using computational biology methods. I am beginning the search for two postdocs now - stay tuned for more details! ewaldlab.org
- Thinking about implementing the Cell Painting assay? Here’s a guide to how it can help in Drug Discovery! t.co/hW1jOhAg7q
- [Not loaded yet]
- [Not loaded yet]
- Yup I just realized what happened and deleted the post before your reply, the organizers of an event said 3pm BST so I searched BST time!
- [Not loaded yet]
- [Not loaded yet]
- Done 🥳
- 212e, I won’t miss you! Not at all!
- When the size of test data is 5 compounds and accuracy is 100% 😌
- We’re thrilled to share our Nature Methods review on #CellPainting — your go-to resource for mastering Cell Painting! From MOA and toxicity prediction to contributing to candidate drugs in clinical trials, we’ve distilled 10 years of insights!
-
View full thread
- I’m writing a paper on showing case by case where this is true and where it’s probably not enough. As far as I know, there is no systematic study on the differences with respect to protein targets, all studies are confounded by lots of viability assays—assays where simple cell counts suffice!
- This ofcourse doesn’t hold for the large datasets where rxrx must have seen improvement (or similar signal with BF images), which makes the case for the switch. In the public domain, we don’t see much of these.. especially where studies only involve around a 100 compounds often—many inactive.
- Applications that are changing the game: Predicting MoAs: By comparing unknown compounds to landmark ones. Toxicity profiling: From mitochondrial health to drug-induced liver injury. Disease modeling: Creating phenotypic signatures for rare diseases, ...among others
- What excites you most about the future of Cell Painting? Share your thoughts below!
- What is Cell Painting? It's a microscopy-based, high-content assay that uses six fluorescent dyes to "paint" cells, capturing thousands of features like size, shape, texture, and intensity. The goal? Reveal cellular responses to perturbations in a target-agnostic way.
- Why it matters: Cell Painting unlocks phenotypic profiling, helping researchers understand drug mechanisms of action (MoA), predict toxicity, and even design safer, more effective compounds. It's cost-effective, scalable, and widely accessible.
- Here's a Cheminformatics Starter pack! Let me know if you would like to be added! And more importantly, do encourage people to join the blue skies :Dat://did:plc:jbu62p2segkzvnuk3e4h5ga7/app.bsky.graph.starterpack/3lb4lirxegh2d
- [Not loaded yet]
- Done! Do share with others !
- The JUMP-Cell Painting findings with genetic perturbations are now released!
- Now on biorxiv! The JUMP-Cell Painting Consortium’s paper: “Morphological map of under- and over-expression of genes in human cells” This is the genetic perturbation portion of the JUMP’s dataset; our chemical perturbation paper will come in a few months www.biorxiv.org/content/10.1...
- Join the Cheminformatics, Machine Learning in Chemistry and Drug Discovery Starter Pack!at://did:plc:jbu62p2segkzvnuk3e4h5ga7/app.bsky.graph.starterpack/3lb4lirxegh2d
- [Not loaded yet]
- Added :D
- [Not loaded yet]
- Done! Welcome to 🦋
- [Not loaded yet]
- Absolutely! Do share with others!
- [Not loaded yet]
- Yess! Welcome to 🦋
- [Not loaded yet]
- Added :D
- [Not loaded yet]
- Added
- [Not loaded yet]
- Added :D Do share the link on other platforms! go.bsky.app/FYr1asG
- Will save more time that asking ChatGPT!
- This is pretty dope. Add pylustrator.start() at the top of a script and your plots are interactively scalable, composable, movable, etc with your mouse www.youtube.com/watch?v=xXPI... pylustrator.readthedocs.io/en/latest/ h/t @tommytang.bsky.social
- [Not loaded yet]
- Yes!!
- Do you want to make biological sense of Cell Painting images? Try BioMorph! BioMorph maps CellProfiler features to a new feature space that’s more interpretable and contains the same signal strength. Try it at broad.io/BioMorph !
- [Not loaded yet]
- [Not loaded yet]
- 16. If repost is the only way to make it to following feed, then perhaps making reposting easier—single click while quote repost can be only on holding down the repost button for a few seconds.
- I've been on here for 15 months, and the change over the last three weeks is truly stunning. Literally many thousands of new scientists on #SciSky. Community building is totally different here than the slow accretion method over years on Twitter. Here are a few thoughts from what I see. 1/n
- I would add one more: start posting and replying! The only way momentum is maintained is if ppl are interesting here… don’t be a bystander just eyeing the buffet and not bringing anything to the potluck 😄
- @bsky.app needs to make reposting as simple as likes. Perhaps a one click repost and a hold long for quote repost.
- [Not loaded yet]
-
View full threadYou would want to enable Priority Notifications under Notification Settings. Perhaps that helps with new followers.
- Okay I just verified. Enabling this limits notifications to only events from people you follow
