Aviv Regev’s research while affiliated with Genentech and other places

Understanding the impact of genetic perturbations on cellular behavior is crucial for biological research, but comprehensive experimental mapping remains infeasible. We introduce PRESAGE (Perturbation Response EStimation with Aggregated Gene Embeddings), a simple, modular, and interpretable framework that predicts perturbation-induced expression changes by integrating diverse knowledge sources via gene embeddings. PRESAGE transforms gene embeddings through an attention-based model to predict perturbation expression outcomes. To assess model performance, we introduce a comprehensive evaluation suite with novel functional metrics that move beyond traditional regression tasks, including measures of accuracy in effect size prediction, in identifying perturbations with similar expression profiles (phenocopy), and in prediction of perturbations with the strongest impact on specific gene set scores. PRESAGE outperforms existing methods in both classical regression metrics and our novel functional evaluations. Through ablation studies, we demonstrate that knowledge source selection is more critical for predictive performance than architectural complexity, with cross-system Perturb-seq data providing particularly strong predictive power. We also find that performance saturates quickly with training set size, suggesting that experimental design strategies might benefit from collecting sparse perturbation data across multiple biological systems rather than exhaustive profiling of individual systems. Overall, PRESAGE establishes a robust framework for advancing perturbation response prediction and facilitating the design of targeted biological experiments, significantly improving our ability to predict cellular responses across diverse biological systems.

Human airways contain specialized rare epithelial cells including CFTR-rich ionocytes that regulate airway surface physiology and chemosensory tuft cells that produce asthma-associated inflammatory mediators. Here, using a lung cell atlas of 311,748 single cell RNA-Seq profiles, we identify 687 ionocytes (0.45%). In contrast to prior reports claiming a lack of ionocytes in the small airways, we demonstrate that ionocytes are present in small and large airways in similar proportions. Surprisingly, we find only 3 mature tuft cells (0.002%), and demonstrate that previously annotated tuft-like cells are instead highly replicative progenitor cells. These tuft-ionocyte progenitor (TIP) cells produce ionocytes as a default lineage. However, Type 2 and Type 17 cytokines divert TIP cell lineage in vitro, resulting in the production of mature tuft cells at the expense of ionocyte differentiation. Our dataset thus provides an updated understanding of airway rare cell composition, and further suggests that clinically relevant cytokines may skew the composition of disease-relevant rare cells.

Biomedical research underpins progress in our understanding of human health and disease, drug discovery, and clinical care. However, with the growth of complex lab experiments, large datasets, many analytical tools, and expansive literature, biomedical research is increasingly constrained by repetitive and fragmented workflows that slow discovery and limit innovation, underscoring the need for a fundamentally new way to scale scientific expertise. Here, we introduce Biomni, a general-purpose biomedical AI agent designed to autonomously execute a wide spectrum of research tasks across diverse biomedical subfields. To systematically map the biomedical action space, Biomni first employs an action discovery agent to create the first unified agentic environment -- mining essential tools, databases, and protocols from tens of thousands of publications across 25 biomedical domains. Built on this foundation, Biomni features a generalist agentic architecture that integrates large language model (LLM) reasoning with retrieval-augmented planning and code-based execution, enabling it to dynamically compose and carry out complex biomedical workflows -- entirely without relying on predefined templates or rigid task flows. Systematic benchmarking demonstrates that Biomni achieves strong generalization across heterogeneous biomedical tasks -- including causal gene prioritization, drug repurposing, rare disease diagnosis, microbiome analysis, and molecular cloning -- without any task-specific prompt tuning. Real-world case studies further showcase Biomni's ability to interpret complex, multi-modal biomedical datasets and autonomously generate experimentally testable protocols. Biomni envisions a future where virtual AI biologists operate alongside and augment human scientists to dramatically enhance research productivity, clinical insight, and healthcare. Biomni is ready to use at https://biomni.stanford.edu, and we invite scientists to explore its capabilities, stress-test its limits, and co-create the next era of biomedical discoveries.

Understanding how genetic interventions affect a phenotype is key to revealing causal gene regulatory mechanisms and finding novel drug targets. Pooled, high-content perturbation screening methods like Perturb-seq allow us to assess the impact of each of a large number of genetic interventions on a rich cellular profile of RNA or other features, facilitating such discoveries. However, the overall scale of perturbations, especially when considering combinations of genes for perturbation, cannot be tackled exhaustively in the lab. An alternative is to use an iterative design: By leveraging the modularity and sparsity of gene cir- cuits along with prior biological knowledge, we can predict the impact of unseen genetic perturbations on these profiles and group genes with similar effects into co-functional modules, followed by a new round of Perturb-Seq to test these pre- dictions and improve the overall performance of the model. These iterative cycles of experiment and prediction allow prioritizing genes for testing, maximizing the knowledge gleaned from fixed experimental resources, and opening the way to learn general predictive models. Designing these experiments requires a system that can analyze a cellular system, incorporate new and existing knowledge, use statistical tools, predict the effects of unseen perturbations, and prioritize the set of perturbations for the next iteration. These can be time-consuming tasks for scientists and require multiple different skills. Here, we developed PerTurboAgent, an LLM-based agent that excels in predicting candidate gene panels for iterative Perturb-Seq experiments through self-directed data analysis and knowledge retrieval. We evaluated PerTurboAgent based on its ability to identify genes with a phenotypic impact on gene expression upon perturbation in genome-scale perturbation data. PerTurboAgent outperforms existing agent-based and active learning strategies, offering an efficient and understandable approach to designing sequential perturbation experiments

Reconstruction and joint embedding have emerged as two leading paradigms in Self Supervised Learning (SSL). Reconstruction methods focus on recovering the original sample from a different view in input space. On the other hand, joint embedding methods align the representations of different views in latent space. Both approaches offer compelling advantages, yet practitioners lack clear guidelines for choosing between them. In this work, we unveil the core mechanisms that distinguish each paradigm. By leveraging closed form solutions for both approaches, we precisely characterize how the view generation process, e.g. data augmentation, impacts the learned representations. We then demonstrate that, unlike supervised learning, both SSL paradigms require a minimal alignment between augmentations and irrelevant features to achieve asymptotic optimality with increasing sample size. Our findings indicate that in scenarios where these irrelevant features have a large magnitude, joint embedding methods are preferable because they impose a strictly weaker alignment condition compared to reconstruction based methods. These results not only clarify the trade offs between the two paradigms but also substantiate the empirical success of joint embedding approaches on real world challenging datasets.

Solid tumors are spatially heterogeneous in their genetic, molecular, and cellular composition, but recent spatial profiling studies have mostly charted genetic and RNA variation in tumors separately. To leverage the potential of RNA to identify copy number alterations (CNAs), we develop SlideCNA, a computational tool to extract CNA signals from sparse spatial transcriptomics data with near single cellular resolution. SlideCNA uses expression-aware spatial binning to overcome sparsity limitations while maintaining spatial signal to recover CNA patterns. We test SlideCNA on simulated and real Slide-seq data of (metastatic) breast cancer and demonstrate its potential for spatial subclone detection. Supplementary Information The online version contains supplementary material available at 10.1186/s13059-025-03573-y.