Excited to see our hierarchical cross-entropy strategy for improved cell type annotation out today in @natcomputsci.nature.com! Congratulations to the team! Try out the HCE loss with your own models and datasets:
📄: www.nature.com/articles/s43...
💻: github.com/microsoft/hc...
📢Out now! Sebastiano Cultrera di Montesano, Peter S. Winter, @lcrawford.bsky.social, and colleagues present a hierarchical cross-entropy loss that improves performance of single-cell annotation models. www.nature.com/articles/s43... 🖥️ 🧬
Come do a research internship with us! Details on the posting and how to upload a research statement given below!
"What James Watson got wrong about DNA"
By the great Sohini Ramachandran (@sramach.bsky.social) and your boy for The Boston Globe (@bostonglobe.com).
www.bostonglobe.com/2025/11/14/o...
Alan did a ton of work on our latest revision. Message of the story still holds: more data does not necessarily lead to better downstream performance for many models.
Check out all our new analyses here: www.biorxiv.org/content/10.1...
Shoot me a message and we can figure it out
Hi! What types of non technical roles were you thinking of? And for full time or internships?
Come do an internship with us! Please apply
Gindra, Palla, Nguyen, Wagner, Tran, Theis, Saur, Crawford, Peng: A Large-Scale Benchmark of Cross-Modal Learning for Histology and Gene Expression in Spatial Transcriptomics https://arxiv.org/abs/2508.01490 https://arxiv.org/pdf/2508.01490 https://arxiv.org/html/2508.01490
📣Online now!
📄Sparse modeling of interactions enables fast detection of genome-wide epistasis in biobank-scale studies
🧑🤝🧑 @lcrawford.bsky.social @julian-stamp.bsky.social & co
Do protein language models store different structural elements in factorizable subnetworks?
To find out, we masked out PLM weights to suppress performance on CATH subcategories or secondary structure elements while maintaining performance on other sequences or residues.
The artwork includes a photo of each of the elected with the word Congratulations at the top.
Members have elected Brian Millen as the 122nd president of the association.
Also elected: Julia Sharp, Vice President; Martin Slawski, Council of Sections Representative; Ruixiao Lu, Council of Chapters Representative; Pedro Silva, International Representative. Congratulations! tinyurl.com/ptunrn5z
Pictures of the speakers for the "Leadership in Trustworthy AI" COPSS-NISS Leadership Webinar with David Donoho (Stanford), Michael I. Jordan (UC Berkeley), and Tracy Ke (Harvard) - online Tuesday, April 29, 2025 at 12-1pm ET.
"Leadership in Trustworthy AI" COPSS-NISS Leadership Webinar with David Donoho (Stanford), Michael I. Jordan (UC Berkeley), and Tracy Ke (Harvard) - online Tuesday, April 29, 2025 at 12-1pm ET.
Register at: www.niss.org/events/copss...
Great to see our paper presenting recall, a framework which calibrates clustering for the impact of data "double-dipping" in single-cell studies, out today in AJHG! Congratulations, @alandenadel.bsky.social and co-authors!
New online! Adapting systems biology to address the complexity of human disease in the single-cell era
🔥 Benchmark Alert! MotifBench sets a new standard for evaluating protein design methods in motif scaffolding.
Why does this matter? Reproducibility & fair comparison have been lacking—until now.
Paper: arxiv.org/abs/2502.12479 | Repo: github.com/blt2114/Moti...
A thread ⬇️
🎉Congratulations to the authors!🎉
The list of accepted papers for #RECOMB2025 is now live ➡️ recomb.org/recomb2025/a...
What are your plans for Seoul? Let us know!
www.youtube.com/watch?v=uztj...
#seoul
Quick post to close out the week - in our newest preprint,
@julian-stamp.bsky.social scales the marginal epistasis test to work on biobanks! The key is that using trait-specific information to induce sparsity in the modeled gene-interactions greatly improves both runtime and power
Evaluating the role of pre-training dataset size and diversity on single-cell foundation model performance
Figure 1
Figure 2
Figure 3
Evaluating the role of pre-training dataset size and diversity on single-cell foundation model performance [new]
Rethinking cancer drug synergy prediction: a call for standardization in machine learning applications https://www.biorxiv.org/content/10.1101/2024.12.24.630216v1
📄 Preprint: www.biorxiv.org/content/10.1...
💻 Code: github.com/microsoft/sc...
Joint work with an amazing group of collaborators:
Madeline Hughes
Akshaya Thoutam
@anaygupta.bsky.social
Andrew Navia
@nfusi.bsky.social
Sri Raghavan
Peter Winter
@avapamini.bsky.social
In our newest preprint, we show that simply increasing the size of pre-training datasets doesn't necessarily improve the performance of single-cell foundation models on downstream tasks.
Really proud of @alandenadel.bsky.social for leading this effort! See his thread below for more details👇
This starter pack is in honor of Assistant Professor Antentor Hinton Jr (AJ) at Vanderbilt School of Medicine - Basic Sciences. AJ curated the first 100 Inspiring Black Scientists in America list and in collaboration with the Community of Scholars expanded this list to the 1000.
go.bsky.app/DsJrwR
Thanks for the heads up!
Looking to find your MSR colleagues? I put together a starter pack.
Msg me if I missed you (or you just joined).
go.bsky.app/NxtpELZ
Oh this is super helpful! I’ve been trying to find everyone lol
In the spirit of trying to be more active on here... the deadline for our BioML internship is tomorrow 11/22.
If you are currently a PhD student please consider applying --- all we need is a CV and research statement!
jobs.careers.microsoft.com/global/en/jo...
