--- title: Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci... description: Xaira Therapeutics built X-Cell, a virtual cell model for drug discovery, by first creating X-Atlas — a proprietary dataset generated through massively... canonical: https://daily.dev/posts/xaira-s-x-cell-model-for-drug-discovery-bo-wang-ci-chu-chief-discovery-officer-chief-ai-scient-gxtj6ukmf twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci Chu, Chief Discovery Officer & Chief AI Scientist) | daily.dev og:description: Xaira Therapeutics built X-Cell, a virtual cell model for drug discovery, by first creating X-Atlas — a proprietary dataset generated through massively... og:url: https://daily.dev/posts/xaira-s-x-cell-model-for-drug-discovery-bo-wang-ci-chu-chief-discovery-officer-chief-ai-scient-gxtj6ukmf og:image: https://api.daily.dev/og/posts/gXtj6UkMf.png og:image:alt: Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci Chu, Chief Discovery Officer & Chief AI Scientist) og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci Chu, Chief Discovery Officer & Chief AI Scientist) **[Latent Space](https://daily.dev/sources/latentspace)** · 4 min read · 0 upvotes · 0 comments ## Summary Xaira Therapeutics built X-Cell, a virtual cell model for drug discovery, by first creating X-Atlas — a proprietary dataset generated through massively parallel CRISPR-based experiments that perturb individual genes to establish causal relationships in gene expression. The key insight is that existing datasets like CELLxGENE describe cell states but lack causal signal, causing models to hit an information ceiling around 1.5B parameters. By generating ~30x more information-rich data, Xaira's team unlocked continued scaling with both parameters and compute. The team also switched from autoregression to diffusion-based modeling. Bo Wang (scGPT author) and Ci Chu discuss the data strategy, model architecture choices, and how beating the linear baseline validates their approach to AI-driven drug development. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why do RNA expression models trained on CELLxGENE fail to predict effects of gene changes? Because CELLxGENE-trained models learn correlations between cell types and states rather than causal relationships, and gene expression changes are highly correlated, making it nearly impossible to determine what causes what. Without perturbation data showing what happens when individual genes are deliberately altered, models can describe cell states but cannot reliably predict the effect of a specific gene edit or drug. _Anyone building predictive models for gene editing or drug targets can follow this reasoning on daily.dev._ ### What is X-Atlas and how does it relate to Xaira's X-Cell model? X-Atlas is a causal dataset built from CRISPR-based experiments that run millions of tests in parallel, perturbing genes one at a time to observe upstream and downstream effects. X-Cell is the model trained on this roughly 30x information-richer dataset, which unlike prior models built on CELLxGENE's correlational data, can scale further with added parameters and compute because it isn't limited by information-sparse training data. _Track how causal datasets like this reshape virtual cell modeling by following drug discovery AI on daily.dev._ ### Why did Xaira's team switch from autoregression to diffusion for their virtual cell model? The team abandoned autoregression in favor of diffusion for building X-Cell, a decision discussed alongside justifying a mix of architectural priors and how those choices compare against improvements gained from higher-quality data. Specific tradeoffs between data quality and model architecture were core to the design decisions behind X-Cell's performance over prior linear baselines. _Engineers weighing diffusion versus autoregressive architectures can keep up with cases like this on daily.dev._ ## Similar posts on daily.dev - [Rivercell raises $25M to build an AI model of how cells respond to drugs](https://daily.dev/posts/rivercell-raises-25m-to-build-an-ai-model-of-how-cells-respond-to-drugs-jiiz2yknn) · The Next Web · 0 upvotes · 0 comments - [AI to help researchers see the bigger picture in cell biology](https://daily.dev/posts/ai-to-help-researchers-see-the-bigger-picture-in-cell-biology-cgech0fnd) · MIT News · 0 upvotes · 0 comments - [Exai Bio & Databricks: Accelerating AI-Powered Liquid Biopsy for Early Cancer Detection](https://daily.dev/posts/exai-bio-databricks-accelerating-ai-powered-liquid-biopsy-for-early-cancer-detection-2kj7yic7j) · databricks · 0 upvotes · 0 comments - [Agentic AI framework in life sciences for R&D](https://daily.dev/posts/agentic-ai-framework-in-life-sciences-for-r-d-vcvbbbffj) · Google Cloud · 0 upvotes · 0 comments - [AI is spitting out more potential drugs than ever. This start-up wants to figure out which ones matter.](https://daily.dev/posts/ai-is-spitting-out-more-potential-drugs-than-ever-this-start-up-wants-to-figure-out-which-ones-matt-usqozoajm) · TechCrunch · 0 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#genai](https://daily.dev/tags/genai), [#biotech](https://daily.dev/tags/biotech) [View this post on daily.dev](https://daily.dev/posts/xaira-s-x-cell-model-for-drug-discovery-bo-wang-ci-chu-chief-discovery-officer-chief-ai-scient-gxtj6ukmf) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci Chu, Chief Discovery Officer & Chief AI Scientist)","url":"https://daily.dev/posts/xaira-s-x-cell-model-for-drug-discovery-bo-wang-ci-chu-chief-discovery-officer-chief-ai-scient-gxtj6ukmf","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/xaira-s-x-cell-model-for-drug-discovery-bo-wang-ci-chu-chief-discovery-officer-chief-ai-scient-gxtj6ukmf"},"datePublished":"2026-07-21T19:36:34.693Z","dateModified":"2026-09-13T20:30:16.637Z","description":"Xaira Therapeutics built X-Cell, a virtual cell model for drug discovery, by first creating X-Atlas — a proprietary dataset generated through massively...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0d847f9846be2f3678f7ce5d26dd26f4?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0d847f9846be2f3678f7ce5d26dd26f4?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Latent Space","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Latent Space","logo":"https://media.daily.dev/image/upload/s--DnJ9laFj--/f_auto,q_auto/v1780213196/logos/latentspace?_a=BAMAMiWQ0","url":"https://daily.dev/sources/latentspace"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/xaira-s-x-cell-model-for-drug-discovery-bo-wang-ci-chu-chief-discovery-officer-chief-ai-scient-gxtj6ukmf","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,genai,biotech","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Latent Space","item":"https://daily.dev/sources/latentspace"},{"@type":"ListItem","position":3,"name":"Xaira’s X-Cell model for Drug Discovery (Bo Wang & Ci Chu, Chief Discovery Officer & Chief AI Scientist)"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/xaira-s-x-cell-model-for-drug-discovery-bo-wang-ci-chu-chief-discovery-officer-chief-ai-scient-gxtj6ukmf#faq","mainEntity":[{"@type":"Question","name":"Why do RNA expression models trained on CELLxGENE fail to predict effects of gene changes?","acceptedAnswer":{"@type":"Answer","text":"Because CELLxGENE-trained models learn correlations between cell types and states rather than causal relationships, and gene expression changes are highly correlated, making it nearly impossible to determine what causes what. Without perturbation data showing what happens when individual genes are deliberately altered, models can describe cell states but cannot reliably predict the effect of a specific gene edit or drug. Anyone building predictive models for gene editing or drug targets can follow this reasoning on daily.dev."}},{"@type":"Question","name":"What is X-Atlas and how does it relate to Xaira's X-Cell model?","acceptedAnswer":{"@type":"Answer","text":"X-Atlas is a causal dataset built from CRISPR-based experiments that run millions of tests in parallel, perturbing genes one at a time to observe upstream and downstream effects. X-Cell is the model trained on this roughly 30x information-richer dataset, which unlike prior models built on CELLxGENE's correlational data, can scale further with added parameters and compute because it isn't limited by information-sparse training data. Track how causal datasets like this reshape virtual cell modeling by following drug discovery AI on daily.dev."}},{"@type":"Question","name":"Why did Xaira's team switch from autoregression to diffusion for their virtual cell model?","acceptedAnswer":{"@type":"Answer","text":"The team abandoned autoregression in favor of diffusion for building X-Cell, a decision discussed alongside justifying a mix of architectural priors and how those choices compare against improvements gained from higher-quality data. Specific tradeoffs between data quality and model architecture were core to the design decisions behind X-Cell's performance over prior linear baselines. Engineers weighing diffusion versus autoregressive architectures can keep up with cases like this on daily.dev."}}]} ```