Fabian Joachim Theis

HARP integrates PBMC data from multiple studies to create a reference atlas with consensus annotations and identifies age, sex, and disease-related cellular modules.

This paper introduces LabCompass, a generative inverse-modeling framework that optimizes culture conditions to steer human hematopoietic cell fate in vitro, demonstrating its effectiveness through iterative experimental validation.

ConvexGating automatically learns unbiased, interpretable gating strategies for flow cytometry data, improving consistency and reducing contamination in cell population identification.

AlphaPeptTools provides a scverse-native Python package for analyzing MS-based proteomics data, enabling scalable analysis and integration with other modalities.

OmegAMP generates antimicrobial peptides with controlled physicochemical properties and safety profiles, showing activity against drug-resistant bacteria in vitro and in vivo.

G2T reconstructs tissue geometry from gene expression data using a generative deep learning model, outperforming existing methods on spatial reconstruction benchmarks.

SQUINT learns discrete cell and niche codes from spatial transcriptomics data using residual vector quantization, outperforming baselines in identification and cross-section integration.

The paper maps chemokine landscapes in solid tumors using spatial transcriptomics, identifying conserved and tumor-specific patterns and revealing perivascular CCR7 dendritic cells as key sources of lymphocyte-attracting chemokines.

The paper profiles human myeloid cells across aging and Alzheimer's disease, identifies AD-associated changes, and links GPNMB to neuroprotection via TREM2.

TERRA is a foundation model for spatial transcriptomics that predicts tissue behavior under perturbation and identifies multi-scale biological insights across tissues.

The paper introduces scAtlasTb, a framework for building and optimizing single-cell reference atlases, demonstrating its application on three Human Cell Atlas projects.

The paper introduces an interpretable neural network architecture for genomic prediction, outperforming existing methods in disease risk prediction and capturing non-linear genetic interactions.

Rectangle provides a scalable deconvolution framework for bulk RNA-seq using single-cell data, achieving high accuracy and robustness across diverse benchmarks.

The paper uses temporal single-cell RNA-seq and other methods to study how Salmonella rewires macrophage gene expression and transcription factor activity during infection.

This paper introduces PertOmni, a multimodal framework that aligns transcriptomic data with text and image embeddings to improve perturbation effect prediction.

The paper constructs a healthy human pancreas reference atlas to contextualize disease states and benchmark models.

The paper benchmarks gene expression reconstruction from single-cell latent representations, finding autoencoders perform best at low dimensionality and foundation models depend on decoder architecture.

Cellpin imputes and denoises spatial transcriptomes using a variational autoencoder trained on single-cell RNA-seq, outperforming existing methods without requiring cross-modality alignment.

RegVelo integrates gene regulatory networks with dynamic modeling to predict cellular changes and perturbations, showing improved performance in cell cycle, hematopoiesis, and pancreatic development datasets.

InterScale models both local and global cellular interactions in spatial transcriptomics data, revealing spatially restricted differentiation programs and disease-associated tissue remodeling.

Phoenix predicts spatially resolved single-cell gene expression from histology, identifying new biomarkers and validating them across cancers and treatment regimens.

UniversalEPI predicts cell type-specific and differential chromatin interactions using DNA sequence and ATAC-seq data, showing robust performance across cell types and conditions.

This paper introduces NTVE, a non-destructive transcriptomics method using vesicular export to monitor RNA dynamics in living cells over time.

Human Developmental Cell Atlas integrates ~4.6 million cells/nuclei from 4-22 post-conception weeks into 454 refined cell types and maps 114 spatial niches across 6PCW whole-embryo sections.

TarDis disentangles multiple covariates in single-cell genomics, improving data integration and out-of-distribution predictions by generating structured latent representations.

This study shows TNF-α disrupts the malate-aspartate shuttle in iPSC-derived enteric neural cells from PD patients, causing metabolic shifts and mitochondrial dysfunction, which can be reversed by targeting glutamate metabolism.

This paper develops a continual learning framework to map colorectal cancer cell states to causal mechanisms through integration of single-cell data and perturbation experiments.

This paper generates cross-species single-cell atlases of pancreatic cancer to study progression, therapy response, and immune evasion across human and mouse models.

This paper introduces a metric to evaluate synergistic integration in multimodal cell models and shows when it benefits tasks over unimodal approaches.

The paper presents a longitudinal multimodal benchmarking dataset and competition to evaluate modality prediction methods for gene regulatory inference, showing top models outperform existing approaches.