Kun-Hsing Yu

Associate Professor, Department of Biomedical Informatics, Harvard Medical School

[Kun-Hsing Yu @kunhsingyu]

Many thanks to @cat_caruso for the insightful press release: https://t.co/44D25Vy6tG
@HarvardDBMI @harvardmed @Harvard

[Kun-Hsing Yu @kunhsingyu]

New in @CellRepMed (December cover): Pathology AI can show demographic performance gaps.

We introduce FAIR-Path, which reduces these gaps by ~90% across 20 cancer types without sacrificing accuracy.
Paper: https://t.co/FewdapJEeS

Toward more generalizable AI in medicine.

[Kun-Hsing Yu @kunhsingyu]

Grateful to our outstanding team (2/2):
@ Nancy U. Lin @NeuropathDM @KeithLigon5 @ Ying-Chun Lo @ Nipon Chaisuriya @ David J. Cook @ Adelheid Woehrer @ Jeffrey Meyerhardt @shuji_ogino @MacNasrallah @Jeff_golden1 @ Sabina Signoretti @ Jung-Hsien Chiang

[Kun-Hsing Yu @kunhsingyu]

Grateful to our outstanding team (1/2):
@ Shih-Yen Lin @ Pei-Chen Tsai @ Fang-Yi Su @ Chun-Yen Chen @ Fuchen Li @ Junhan Zhao @ Yuk Yeung Ho @ Tsung-Lu Michael Lee @Liz_Healey @ Po-Jen Lin @ Ting-Wan Kao @ Dmytro Vremenko @ Thomas Roetzer-Pejrimovsky @lmsholl @ Deborah Dillon

[Pranav Rajpurkar @pranavrajpurkar]

Hiring an AI engineer at @a2zradiologyai as we continue to scale the world's best foundation models for radiology AI.
If you've built/shipped large CV models (PyTorch 2.x, multi-GPU), are ruthlessly organized in documentation, and move with high agency. https://t.co/Rm6ByTHhy8

[Pranav Rajpurkar @pranavrajpurkar]

A historic day for radiology AI: a2z Radiology AI is bringing a first-of-its-kind capability to the U.S. market—a single system that triages seven urgent conditions on abdomen-pelvis CT.

Today we're announcing FDA clearance for a2z-Unified-Triage.

[Marinka Zitnik @marinkazitnik]

Enormous thanks to @AdaFang_ @YuanqiD, Mia Rosenfeld @iambic_ai, and the entire organizing team for spearheading this amazing @NeurIPSConf workshop!

Huge congratulations on a stellar run of the 6th AI for Science Workshop @AI_for_Science

[Marinka Zitnik @marinkazitnik]

AI agents for science are featured as a “Method to Watch” in @NatureMethods, https://t.co/5r3ByuFJ1B

The piece also highlights our roadmap outlining capabilities, the need for thorough evaluations, and open challenges for AI agents, https://t.co/jfCUoFZFfW

[Michael Baym @baym]

How the genes plasmids carry affect their hosts has been well studied, but this smaller scale of evolution has proven much tougher despite decades of theory. Plasmids are small and replication is fast and hard to measure. 2/
https://t.co/j3ybjBRZXA

[Isaac Kohane @zakkohane]

The Missing Value of Medical AI. It’s the right hill to take for a safe 21st century medicine @NatureMedicine link 👇

[Isaac Kohane @zakkohane]

Current thinking on alignment @NeurIPSConf 2025: @payal_chandak shared key presentations & I sent these to @google LLMNotebook. Listen for 15 minutes of 10,000 feet overview of considerations that are central to medicine including pluralistic alignment.

[Kun-Hsing Yu @kunhsingyu]

So thankful, @EkaterinaPeshev and @cat_caruso, for the brilliant news release: https://t.co/1tQUcjQGzT @harvardmed

[Kun-Hsing Yu @kunhsingyu]

Delighted to share our new paper! We introduce an uncertainty-aware pathology AI that diagnoses cancers with overlapping profiles and knows when it doesn't know.

Full article: https://t.co/pWUnojRFtU

@HarvardDBMI @NIH @NIGMS @CDMRP @AmericanCancer #DigitalPathology #AI

[Kun-Hsing Yu @kunhsingyu]

@HarvardDBMI @NIH @CDMRP @AmericanCancer Deeply appreciative of our incredible team (2/2):
@KeithLigon5 @OmarArnaout12 @ Thomas Roetzer-Pejrimovsky @ Shih-Chieh Lin @ Natalie NC Shih @ Nipon Chaisuriya @ David J. Cook @ Jung-Hsien Chiang @ Chia-Jen Liu @ Adelheid Woehrer @Jeff_golden1 @MacNasrallah

[Kun-Hsing Yu @kunhsingyu]

Delighted to share our new paper! We introduce an uncertainty-aware pathology AI that diagnoses cancers with overlapping profiles and knows when it doesn't know.

Full article: https://t.co/pWUnojRFtU

@HarvardDBMI @NIH @NIGMS @CDMRP @AmericanCancer #DigitalPathology #AI

[Kun-Hsing Yu @kunhsingyu]

@HarvardDBMI @NIH @CDMRP @AmericanCancer Deeply appreciative of our incredible team (1/2):
@ Junhan Zhao @ Shih-Yen Lin @ Raphaël Attias @lizamathewskim @ Christian Engel @ Guillaume Larghero @ Dmytro Vremenko @ Ting-Wan Kao @ Tsung-Hua Lee @ Yu-Hsuan Wang @ Cheng Che Tsai @ElianaMarostica @ Ying-Chun Lo @NeuropathDM

[Isaac Kohane @zakkohane]

Impressive but skynet™ will remember this unkindness.

[Pranav Rajpurkar @pranavrajpurkar]

Delighted to publish our review article on the generative era of medical AI in Cell!
w/ @EricTopol JohnFahrner and Emma Chen

[Pranav Rajpurkar @pranavrajpurkar]

Fun to share our work on Generalist Medical AI, and our applications in GI explored with Dr Tyler Berzin and Romain Hardy

[Marinka Zitnik @marinkazitnik]

Ever wish you could hit "undo" on disease? 🩺🔄

https://t.co/gAR73QFOtC

Most drug discovery asks: what does this perturbation do to cells? But we can also ask the reverse: which perturbations undo a disease signature and move cells back toward health?

That's the idea behind PDGrapher, just published in Nature Biomedical Engineering @natBME @justguadaa @mmbronstein

⚠️ Challenge
Phenotype-based discovery screens compounds forward, measuring whether they improve disease phenotypes
But it does not address which perturbations will reverse disease signatures or capture combinatorial targets that may be essential for restoring health

💡 Approach
PDGrapher tackles perturbagen discovery: finding the genes, drugs, or combinations that can shift a diseased cell state toward health

It represents disease cell states on molecular networks and asks: which perturbations will move these states toward health?

It uses a causally inspired model that overlays gene-expression signatures onto proxy causal graphs. Perturbagens are encoded as edge modifications, allowing the model to reason about how interventions reshape cellular processes

It is trained with two objectives:
💡 Cycle-consistency loss: ensures predicted perturbations move a diseased state toward its observed treated state
💡 Supervision loss: directly aligns predictions with experimental treatment outcomes

[Marinka Zitnik @marinkazitnik]

Update from CURE-Bench @NeurIPSConf: 900+ entrants and 1,100+ submissions

👉 New teams welcome. Tracks, rules, and leaderboard: https://t.co/95IwtDNNhk #NeurIPS2025 #CUREBench

💻 Starter kit https://t.co/dMC9hrrSCe now includes code + tutorials for @OpenAI GPT-OSS open-weight reasoning models. Not yet saturated by recent models. Add your model and see how it stacks up! Thanks to @Reza_Shamji

Shout out to @GaoShanghua Sufian Aldogom @RichardYXZhu @ZKong50693 @xiaorui_su @CurtGinder

Huge thanks to our partners: @cziscience (Rare as One program) @MilkenInstitute @BiswasFamilyFdn @harvardmed @BrighamWomens @HarvardDBMI @KempnerInst @MIT @broadinstitute