
Eric & Wendy Schmidt postdoctoral fellow, co-mentored by Van Allen @DanaFarber and Uhler @BroadInstitute / PhD in Computational Oncology, ETH Zürich
Joined February 2022
- Tweets175
- Following152
- Followers183
- Likes158
Pinned Tweet
🧵 Introducing SpatialFusion: a lightweight #multimodal #FoundationModel for pathway-informed spatial #niche mapping.
📄 doi.org/10.64898/2026.03.16.…
It combines #histopathology and #SpatialTranscriptomics to identify functional microenvironments in tissue.
Here’s what we found 👇
Josephine Yates retweeted
1/ Most genes and variants matter somewhere but look silent in any one cell model. Can we build conditions that reveal phenotypes? With @kanishkadey, we engineer cell states to expose hidden disease biology and trace how genetic risk reaches disease genes. biorxiv.org/content/10.64898…
Josephine Yates retweeted
Please welcome our preprint on finding spatial cell arrangements linked to patient phenotypes!
Link: biorxiv.org/content/10.64898…
GRASS-MIL can use regression, classification, or survival labels as targets and provides several interpretability tools to understand what drives its predictions.
Technically, it combines, in an end-to-end model, graph neural networks with multiple-instance learning: a GNN learns representations of local cellular neighborhoods, while attention-based MIL learns which neighborhoods matter for the patient-level phenotype. Thus, predictions can be decomposed into signed contributions from individual neighborhoods, helping identify and characterize phenotype-associated spatial niches.
We demonstrate the GRASS-MIL's efficiency and utility on simulated data and ovarian and lung cancer spatial datasets, including joint analysis across CosMx, Xenium, and MERFISH for the ovarian cohort. The work was led by Lovro Rabuzin @Rabuzin_L , with contributions and input from Josephine Yates @JoYatesResearch , Mina Xu @MinaXu7 , Martina Haberecker, Raphael Werner, Mayura Meerang, Rong Fan @RongFan8 , and Archibald Enninful.
So proud to have been part of this tremendous effort! Congrats Aayush & Valentina on getting this work out.
Excited to share that UniversalEPI is officially published in @OxfordJournals Nucleic Acids Research! 🚀
Check out the final paper and use the model here:
📄 Paper: academic.oup.com/nar/article…
💻 Code: github.com/BoevaLab/Universa…
(1/3) 🧵
1
199
Josephine Yates retweeted
🎉Congratulations to our symposium poster and short talk winners!
🥇First: @AdaFang_ (poster 34)
🥈Second: @AnushChP (poster 24)
🥉Third: @JoYatesResearch (poster 13)
🥉Third: Mahshid Heidari, Mina Karimpour (poster 42)
🏆 Best Short Talk: Maile Hirschmann, @dawnchenx
Josephine Yates retweeted
We’re excited to share that our @Schmidt_Center ALS Target Team is participating in Stage 1 of the @Longitude_Prize on ALS! bit.ly/4tCWpcH @mndassoc @Challenge_Works
Josephine Yates retweeted
Congratulations, @JoYatesResearch!
Honored to be named to the Forbes 30 Under 30 Europe (#ForbesUnder30) list in Science & Healthcare.
Grateful for the incredible mentors and colleagues who’ve supported me along the way.
forbes.com/profile/josephine…
Honored to be named to the Forbes 30 Under 30 Europe (#ForbesUnder30) list in Science & Healthcare.
Grateful for the incredible mentors and colleagues who’ve supported me along the way.
forbes.com/profile/josephine…
A special thanks to @val_boeva @VanAllenLab @CarolineUhler for their invaluable mentorship and @ETH_en @CSatETH @Schmidt_Center @broadinstitute @DanaFarberNews for their support.
5
180
Josephine Yates retweeted
🎉 We are excited to share new work led by @Schmidt_Center postdoctoral fellow @JoYatesResearch: SpatialFusion, a multimodal foundation model that identifies tissue niches based on pathway activity—not just spatial proximity. Learn more ⬇️
@broadinstitute
🧵 Introducing SpatialFusion: a lightweight #multimodal #FoundationModel for pathway-informed spatial #niche mapping.
📄 doi.org/10.64898/2026.03.16.…
It combines #histopathology and #SpatialTranscriptomics to identify functional microenvironments in tissue.
Here’s what we found 👇
SpatialFusion is one step in that direction.
By combining histology, gene expression, signaling, and spatial context, it can reveal tissue states that are invisible to single data types.
1
2
183
Huge thank you to my co-mentors @VanAllenLab and @CarolineUhler for their guidance on this work and for kickstarting my postdoc.
Many thanks as well to my co-authors @MShavakhii and @DrChoueiri, and to the @Schmidt_Center for supporting this research. @broadinstitute @DanaFarber
2
5
227











