Jure Leskovec

AutoScreen improves gene prioritization in functional genomics by integrating AI agents for pre-screen design and post-screen analysis, achieving higher hit recovery and precision than existing methods.

State, a set-based transformer trained on over 100 million perturbed cells, predicts perturbation responses in new cellular contexts better than mean, linear and deep learning baselines.

This study identifies conserved and clade-specific hypothalamic cell types across vertebrates using single-cell transcriptomics, revealing conserved transcription factors and rewired regulatory linkages.

Perturb-ME combines CRISPR screening and single-cell profiling to identify regulatory modules governing gene programs in melanoma.

BiomniBench evaluates LLM agents' research processes in biomedical tasks using expert rubrics, revealing performance gaps in method selection and reasoning.

Bio-BLIP is a multimodal architecture that integrates DNA, genes, proteins, and text to improve variant interpretation without task-specific fine-tuning, achieving better performance in variant prioritization and target gene prediction.

This paper introduces TxConformal, a framework that controls false discoveries in AI-driven therapeutic discovery by adjusting for distribution shifts and providing confidence measures for AI scores.

This paper argues for new benchmarks to evaluate AI models' utility in scientific discovery and introduces PerturbHD as an evaluation framework for AI-enabled hit discovery.

PULSAR is a foundation model that integrates multi-scale biological data to predict disease and clinical outcomes in the human immune system.

This paper introduces a deep reinforcement learning framework for designing new antibiotics, producing potent peptides with broad-spectrum activity against bacteria.

The paper shows that current cell embedding evaluation metrics can be misleading and introduces a new metric, scGraph, to detect distortions in embeddings.

An LLM agent with a literature-mined environment of 150 tools, 105 packages and 59 databases plans in code, matching reported human accuracy on database and sequence question answering.

SpatialAgent integrates large language models with dynamic tool retrieval, multimodal interpretation, and claim verification to automate end-to-end spatial biology workflows from gene-panel design to tissue annotation and hypothesis generation.

KGWAS improves GWAS power in small cohorts by integrating a functional knowledge graph, discovering more associations for rare diseases.

The paper introduces a method to control cell fates by modulating multiple genes using CRISPR tools and machine learning, demonstrating its effectiveness in stem cell differentiation.

SATURN integrates single-cell RNA-seq datasets across species using deep learning to create universal cell embeddings, enabling cross-species analysis and annotation transfer.

The paper presents an aging Drosophila cell atlas, identifies cell type-specific aging patterns, and shows ribosomal gene expression as a conserved aging predictor.

The paper reanalyzes millions of molecular profiles to identify aging interventions using AI agents, finding over 500 interventions that reduce biological age and revealing patterns in aging modulation.

STELLAR annotates spatially resolved single-cell data using geometric deep learning, discovering cell types and states while transferring annotations across tissues and donors.

GEARS predicts transcriptional responses to single and multi-gene perturbations using single-cell RNA-sequencing data and geometric deep learning, outperforming existing methods in identifying genetic interactions.

The paper creates a single-cell transcriptomic atlas of the adult fruit fly, identifying over 250 cell types and analyzing gene signatures and sexual dimorphism across tissues.

The paper maps cell organization and function across human intestine regions using single nuclear RNA and open chromatin assays, revealing distinct cell neighborhoods and regulatory differences.

This paper introduces a multiscale interactome framework to explain drug mechanisms by integrating drugs, diseases, proteins, and functions, showing drugs often target functionally related proteins rather than disease-associated ones.

The paper generates single-cell transcriptomes of Drosophila olfactory receptor neurons across development and adulthood, revealing stable receptor expression and cell-type-specific transcriptomic features.

The paper analyzes single-cell transcriptomes of Drosophila olfactory projection neurons across developmental stages to identify transcriptomic changes and cellular diversity patterns.

The paper identifies superhub cells in neural networks that, when targeted, more effectively stabilize epileptic networks compared to traditional hub cells.

MARS identifies and annotates known and novel cell types in single-cell RNA-seq data by transferring latent representations across datasets and generating interpretable names for new cell types.

OnClass classifies single-cell data into Cell Ontology terms and identifies marker genes for all categories, even those absent in training data.

The paper studies interactome resilience across 1,840 species, finding that interactomes become more resilient over evolution, correlating with environmental adaptability.

Network Enhancement improves the signal-to-noise ratio of weighted biological networks, enhancing downstream analysis performance across multiple applications.