Caroline Uhler

StriMap models TCR-peptide-HLA interactions using structural and sequence features, achieving state-of-the-art performance and identifying disease-associated peptides in ankylosing spondylitis.

TissueFormer is a foundation model that predicts spatial gene expression from histology images at single-cell resolution and scales to diagnostic tasks.

The paper presents a machine learning challenge to predict gene perturbation effects on T cell differentiation, finding that feature representation from perturbational data improves performance and validating top genes experimentally.

SpatialFusion identifies spatial niches through pathway activation patterns, revealing pre-malignant and malignant microenvironments in cancer tissues.

APOLLO learns shared and modality-specific information from multi-modal single-cell data to improve understanding of cell state and regulatory mechanisms.

This paper introduces SCCVAE, a hybrid model combining mechanistic causal models with deep learning to predict and simulate genetic perturbation responses in single-cell data.

This study identifies IL13 as a survival factor for Hodgkin lymphoma cells by analyzing the tumor microenvironment with spatial and single-cell resolution.

This study uses single-cell image-based CRISPR screens to identify host factors regulating Ebola virus infection, revealing new therapeutic targets like UQCRB and STRAP.

MORPH predicts cellular responses to genetic perturbations across data modalities and conditions using a variational autoencoder and attention mechanism.

This paper introduces PUPS, a method that predicts protein subcellular localization in single cells using protein sequences and cellular images, showing generalization to unseen proteins and cell lines.

DSMBind predicts binding energy and designs nanobodies without experimental data, achieving performance comparable to supervised models.

The paper develops a pipeline using Recursive Feature Machines to identify synthetic lethal gene pairs from CRISPR viability data, improving accuracy over prior methods and discovering new candidates.

The paper uses Perturb-seq to systematically study 1,130 E3 ligases and their roles in inflammatory responses in dendritic cells, revealing co-functional modules and genetic interactions.

This paper develops a cross-modal autoencoder framework to integrate cardiac MRI and ECG data, improving phenotype prediction and enabling MRI imputation from ECGs.

The paper introduces DCI, an algorithm to infer differences in gene regulatory networks between conditions using gene expression data, and validates it on bulk and single-cell RNA-seq data.

This paper introduces an autoencoder-based method to integrate single-cell imaging and sequencing data, identifying sub-populations of human naive CD4+ T-cells.

This paper introduces ImageAEOT, a method combining autoencoders and optimal transport to infer single-cell trajectories from imaging data, validated on chromatin images of cancer cells and tissues.