GenBio AI

🧬 We’re at #ICML2025 with five papers across two key workshops on generative AI and biology. Our work spans multimodal foundation models, protein design, spatial biology, and cellular response modeling. 🔬 Highlights of our research: 1. Rapid and Reproducible Multimodal Biological Foundation Model Development with AIDO.ModelGenerator, by Caleb Ellington, et al. A toolkit that simplifies the development and benchmarking of multimodal biological foundation models. 🔗 Read the paper: https://x.genbio.ai/mcaWE4 2. AIDO.Tissue: Spatial Cell-Guided Pretraining for Scalable Spatial Transcriptomics Foundation Model, by Jing Gong, et al. Spatially-aware pretraining that improves downstream performance in tissue-level tasks. 🔗 Read the paper: https://x.genbio.ai/afoBY0 3. Multimodal Benchmarking of Foundation Model Representations for Cellular Perturbation Response Prediction, by Euxhen Hasanaj, et al. The first pan-modal benchmark study evaluating how biological FMs perform in predicting cellular drug and gene responses. 🔗 Read the paper: https://x.genbio.ai/klB2p8 4. Retrieval-Augmented Protein Language Models for Protein Structure Prediction, by Pan Li, et al. Retrieval-enhanced language models that outperform AlphaFold2 in low-MSA scenarios 🔗 Read the paper: https://x.genbio.ai/2v2HO4 5. Uncertainty-Aware Discrete Diffusion Improves Protein Design, by Sazan Mahbub, et al. An uncertainty-aware discrete denoising diffusion model for protein inverse folding. 🔗 Read the paper: https://x.genbio.ai/jRT2kp Workshops featuring Co-Founder and Chief Scientist Eric Xing: ICML 2025 Workshop on Generative AI and Biology 🗓️ July 18, 2025 | 📍 Vancouver Convention Center, East Exhibition Hall A • Keynote Talk: 3:30–4:10 PM PT • Session Panel: 4:10–5:40 PM PT ICML 2025 Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences 🗓️ July 19, 2025 | 📍 Vancouver Convention Center, Meeting Rooms 301–305 • Keynote Talk: 8:40–9:05 AM PT 🧬 Full recap and schedule: https://lnkd.in/eT3VMzm9

Good software should be fast, reliable, reusable, and maintainable. A lot of BioML benchmarking is uh… not. But biology doesn’t standardize to a few data types like language, audio, or images. We’re constantly inventing new ways to measure life. At GenBio AI, it was obvious from day 1 that we couldn’t rely on universal datatypes for benchmarking. We needed new tools to build, test, and productionize the multi-scale and multi-modal biological simulator we’re building. This weekend, I'll be giving a spotlight presentation at the ICML Generative AI in Biology workshop on our approach, AIDO.ModelGenerator. https://lnkd.in/e3kKUb5b We designed AIDO.ModelGenerator to answer 4 core research questions in BioML: 1. What model is best for my use-case? Will pretrained models be helpful? 2. How can pretrained models be efficiently adapted for new tasks? 3. How can we make pretrained models easier to use? 4. How can I apply pretrained models to my data? while being low/no-code, highly hackable, scalable with available hardware, and fully reproducible. Now, the package supports 30+ foundation models, including the AIDO series and popular open-source models like Enformer, ESM, and scFoundation, as well as 300+ datasets from dozens of benchmarks spanning DNA, RNA, protein, cell, and tissue data modalities, with more coming soon. Check out the full list here: https://lnkd.in/eztZRJjn If you're at ICML and interested in building biological simulators and tools to accelerate research, drop me a line (or our all-star team Shuxian Zou, Elijah Cole, Ning Sun, Sohan Addagudi, Le Song, Ziv Bar-Joseph, Eric Xing) and let's chat!

🚨 New paper alert! 🤔 What if protein design models knew where they’re uncertain and adapted their predictions accordingly? 🚀 We introduce uncertainty-aware discrete diffusion for structure-conditioned protein design: a modular framework that enhances sequence generation by signaling where and when to denoise (📄 Preprint: https://lnkd.in/g6rzk_Yc). ⚙️ Guidance through prior-posterior uncertainty signaling 🧬 Protein LLM + structure encoder priors 🏆 SoTA results on inverse folding benchmarks 📍 Catch us at #ICML2025 FM4LS Workshop this week! Deeply grateful to my co-authors and supervisors at GenBio AI: Christoph Feinauer, Caleb Ellington, Le Song, Eric Xing. #DiscreteDiffusion #UncertaintyEstimation #GenerativeModels #ProteinDesign

We will be presenting our paper on Multimodal Benchmarking of Foundation Model Representations for Cellular Perturbation Response Prediction (https://lnkd.in/dYjeq6Zi) at two #ICML2025 Workshops this week: FM4LS (https://fm4ls.github.io/) and Generative AI and Biology (https://lnkd.in/dVJaeX6h). We systematically benchmarked perturbation embeddings across modalities (expression, protein, DNA, prior knowledge, networks). Embeddings based on network and prior knowledge consistently outperformed expression-based FMs, suggesting that structured biology remains a strong foundation for perturbation modeling. #SingleCell #FoundationModels #PerturbationModeling

We're #hiring a new Research Scientist (AI) – Biomedical Imaging in Paris, Île-de-France. Apply today or share this post with your network.

We're #hiring a new Research Scientist (AI) - Cell & Tissue Modeling in Paris, Île-de-France. Apply today or share this post with your network.

I couldn't be more excited to represent GenBio AI at [ICML] Int'l Conference on Machine Learning next week in Vancouver! Our team will be presenting 5 papers highlighting recent progress in building the world's first AI-Driven Digital Organism. Our Co-Founder and Chief Scientist, Eric Xing will be delivering Keynote talks at the following workshops: • ICML 2025 Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences • ICML 2025 Workshop on Generative AI and Biology If you will be attending and are interested in learning more about our work, please reach out to me ahead of time. It would be great to set up a conversation between sessions. See you there!

New from the Foundation Models for Biology Seminar Series, Kexin Huang from Stanford University presents Biomni, a general-purpose biomedical AI agent that automates scientific workflows across domains. #Biomni integrates large language model reasoning, retrieval-based planning, and code execution to complete research tasks like gene prioritization, drug repurposing, rare disease diagnosis, and more. The system constructs a unified environment using tools, datasets, and protocols mined from 25 biomedical domains, enabling broad generalization without task-specific tuning. 🔬 Case studies show Biomni designing full experimental protocols from minimal prompts. 🧰 Tasks span genomics, microbiome analysis, molecular cloning, and beyond. 🎙️ Hosted by Caleb Ellington and Shahin Mohammadi 📺 Watch the talk: https://lnkd.in/eirRr26f 📄 Paper: https://lnkd.in/ebc8S2KP 🌎 Learn more about Biomni: https://lnkd.in/e-hNdtJX 📖 Read the summary blog: https://lnkd.in/eQJ3XF65

We are excited to organize NeurIPS 2025 2nd Workshop on Multi-modal Foundation Models and Large Language Models for Life Sciences. The workshop features a stellar lineup of invited speakers, including Ziv Bar-Joseph, Charlotte Bunne, Simona Cristea, Stefano Ermon, Quanquan Gu, Nilah Monnier Ioannidis, Emma Lundberg, Caroline Uhler, Eric Xing Huge thanks to my co-organizers: James Zou, Le Song, Ruishan Liu, Li Zhang, Aidong Zhang, Eran Segal, Wei Wang Paper submission details will be announced soon. See you in San Diego!

We're #hiring a new Research Scientist (AI) - Cell & Tissue Modeling in Paris, Île-de-France. Apply today or share this post with your network.