←
Eric Sun81 posts
Eric Sun avatar

Eric Sun

@EricDSun

Machine learning for aging, spatial/single-cell omics, immunology || Assistant Professor @MIT | PhD @Stanford

◷ Joined October 2021▧ 20 Photos and videos
Eric Sun avatar
Eric Sun@EricDSun·9 Jun 2025

I’m very excited to be joining MIT Biological Engineering @MITdeptofBE as an Assistant Professor and the Ragon Institute @ragoninstitute as a member in January 2026! I will be recruiting students and postdocs (see more info below).

Post attachment
James Zou avatar
⟳ Eric Sun retweeted
James Zou@james_y_zou·Mar 17

#CellVoyager is published in @NatureMethods! 🔬 Autonomous comp bio agent 🧬 Analyzes bio data to make new discoveries 💻 Runs in Jupyter notebooks → interactive & reproducible 📊 Generated expert-validated insights on COVID-19, cell–cell communication and aging

0:20
Claire Bedbrook avatar
⟳ Eric Sun retweeted
Claire Bedbrook@clairebedbrook·Mar 12

Aging may feel gradual… but what if it’s not? In our paper out today, we tracked fish continuously from puberty until death. This gave us a unique view of how aging unfolds across the adult lifespan. 🧵

0:05
Mert Yuksekgonul avatar
⟳ Eric Sun retweeted
Mert Yuksekgonul@mertyuksekgonul·Jan 22

How to get AI to make discoveries on open scientific problems? Most methods just improve the prompt with more attempts. But the AI itself doesn't improve. With test-time training, AI can continue to learn on the problem it’s trying to solve: test-time-training.github.io…

Post attachment
James Zou avatar
⟳ Eric Sun retweeted
James Zou@james_y_zou·17 Dec 2025

We recently organized #Agents4Science, the 1st conference where LLMs are both authors and reviewers🤖 It was an open experiment to assess how well AI can lead research and review papers. Today we report what we learned in @NatureBiotech Highlights in 🧵

Post attachment
Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

Our preprint can be accessed here: biorxiv.org/content/10.64898…. The API and analysis code for SpatialProp can be found at: github.com/abuendia/spatial-….

Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

On this benchmark, SpatialProp predictions were generally enriched for the causal/forward interaction compared to simple baselines. The same framework can be extended to other known cell-cell interactions and used to evaluate future spatial perturbation models.

Post attachment
Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

(3) How to assess for causal predictions, especially for complex perturbations? We introduce CausalInteractionBench, which comprises sender-receiver gene lists, and examined the enrichment for forward (sender-to-receiver) over reverse (receiver-to-sender) perturbation responses.

Post attachment
Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

(2) SpatialProp can also predict fine-grained changes in center cell gene expression when “steering” the microenvironment towards a target state even when conditioned on the same cell type.

Post attachment
Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

(1) SpatialProp can predict the gene expression of center cells from surrounding cells across multiple datasets and is better than simple baselines (note: SpatialProp is trained independently on each dataset).

Post attachment
Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

We believe that spatial perturbation models are useful if they can: (1) predict gene expression accurately; (2) predict fine-grained cellular changes from microenvironmental perturbation; (3) predict causal/directional spatial perturbation effects.

Post attachment
Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

The SpatialProp pipeline is simple, involving a GNN for predicting gene expression from the spatial microenvironment and a series of post-processing steps (SparseRenorm) to calibrate for prediction errors in the perturbation context.

Eric Sun avatar
Eric Sun@EricDSun·3 Dec 2025

Can we predict the spatial effects of single-cell perturbations? Excited to share our preprint introducing SpatialProp, which computationally propagates single-cell transcriptomic perturbations across tissues, along with key frameworks for evaluating spatial perturbation models.

Post attachment