NVIDIA Healthcare

(2/2) 1⃣ 33x faster image generation MAISI-v2 cuts inference from 1,000 steps to just 30 - whole-body coverage, any anatomy, one model. 2⃣ The world's largest open-source brain MRI dataset MR-RATE: 700K+ volumes from 83K+ patients, paired with radiology reports. Now powering NV-Generate-MR-Brain. 3⃣ One command to get started Open-source code, pretrained weights, royalty-free on RTX GPUs. Plug into your pipeline and go. Already used by researchers for lung cancer classification, prostate lesion detection, MR-to-CT synthesis & brain tumor generation. 🔗 Get started nvda.ws/4v5mFNj

(1/2) 🚨 Data scarcity is the #1 blocker in medical imaging AI. We built the open-source fix. NV-Generate-CTMR synthesizes realistic 3D CT & MRI volumes at scale - with paired segmentation masks - so you can train more robust models without touching real patient data.

👏 Congrats to Genesis Molecular AI and Incyte - their collaboration unlocks new value from pharma data to power high-impact drug discovery. Their collaboration accelerates training and fine-tuning of leading foundation models for small-molecule-protein structure prediction, including Pearl built by Genesis and NVIDIA.

@mehrzadsamadi (3/3) Read their work on @GenomeBiology nvda.ws/4nwvInW

(2/3) 1️⃣ Hybrid Architecture: It combines fast transcriptome-guided mapping with a Transformer-based model to resolve difficult splice junctions 2️⃣ High Accuracy: The model hits over 97% accuracy on splice-junction predictions in its benchmark evaluations 3️⃣ Beating the Baselines: It outperforms several widely used aligners, proving that ML adds the most value when targeted at specific bottlenecks

(1/3) 🚨New RNA-seq alignment preprint The team at NVIDIA (Fadel Berakdar, Tong Zhu, @mehrzadsamadi, Pankaj Vats) and Genentech (Thomas D. Wu) just showed how targeted ML can upgrade traditional genomics pipelines with DeepSAP.

(3/3) 📚 nvda.ws/4tXxeBO Based on the 2026 @AnnualReviews of Biomedical Engineering piece ‘Federated Learning in Healthcare: From Research to Real-World Deployment’ by @SpyridonBakas, @XiaoxiaoLi8, Prashant Shah, and Holger R. Roth.

(2/3) 🏥 Train shared models across hospitals without sharing raw patient data 🌐 Tackle non-iid data, system heterogeneity, and robust aggregation across sites 🔐 Add secure aggregation + differential privacy because FL alone ≠ guaranteed privacy 📊 Learn from real deployments in imaging, EHRs, and outcome prediction, not just simulations 🛠️ Treat governance, infra, and MLOps for FL as core research problems, not afterthoughts

(1/3) We don’t need more data in one place - we need better ways to learn from data that can’t move. That’s where federated learning in healthcare is finally getting real.

Grateful to @NVIDIAAI for early access to Nemotron 3 Nano Omni. Healthcare data is everywhere; rarely connected. Exploring Agentic multimodal AI to unify text, audio, images & video for India-scale care. Read More: bit.ly/4cNRvUB #EkaCare #NVIDIA #NemotronOmni

(3/3) Try KERMT today nvda.ws/4mWsZDY Check out the challenge 👇 nvda.ws/4eiyURv

(2/3) Why KERMT? ✅ Pretraining: Uses Kinetic GROVER for rich structural/chemical embeddings ✅ Efficiency: Streamlined for NVIDIA-accelerated multi-task learning ✅ Reproducibility: Fully open-source and ready for your next lead-op pipeline

(1/3) 🛠️ We designed KERMT to bridge the gap between self-supervised graph learning and industrial ADMET workflows. Recent results from the OpenADMET-ExpansionRX challenge confirm the impact: 5 of the top 20 models used KERMT as their backbone to predict critical pharmacokinetic endpoints.

Scaling laws are powering AI. It’s time to scale biology. Today we’re launching the Virtual Biology Initiative to generate the data to unlock scaling laws in biology and build accurate predictive models of the cell. Digital representations of proteins are already expanding our understanding of life at the molecular level, and accelerating the design of molecules and medicines. Accurate digital representations of the cell could reveal the mechanisms that are responsible for disease, and show how to reverse them. The protein data bank, and worldwide repositories of protein sequence biodiversity were created through decades of work by the scientific community. The advances in artificial intelligence for proteins would not have been possible without them. The cell is orders of magnitude more complex, and we will need to create the data in just a few years rather than decades. This will require a coordinated global effort. We're partnering with Broad, Wellcome Sanger, Arc, Allen, Human Cell Atlas, Human Protein Atlas, NVIDIA, and Renaissance Philanthropy. Biohub is contributing to this effort as both a funder and a builder. We are developing microscopy to observe millions of cells in living organisms, and cryo-ET to resolve the cell in atomic detail. We're building instruments that expand the range of modalities and parameters that can be simultaneously measured. We’re developing molecular, cellular, and tissue engineering to create models of disease and design interventions. The data we generate will be available to the worldwide scientific community. We’re also committing $100M over the next five years to support work beyond Biohub. We invite other scientific teams and funders to join. Link: biohub.org/news/virtual-b…

(2/2) You can now run Boltz model structure predictions on tens of thousands of tokens (e.g., >20k) by scaling across many GPUs - overcoming single-device memory limits without approximations or chunking. Full write-up, including distributed implementations of triangle operations 👉 nvda.ws/4tFWuN5

(1/2) Struggling with the memory limits of structure prediction for large biomolecular systems? ✂️ Common workarounds, like cropping sequences or chunking inputs, can break global interactions and bias predictions by removing long-range context. Our latest tech blog explains how the NVIDIA BioNeMo team implemented Context Parallelism (FoldCP): distributing a single large molecular system across multiple GPUs, rather than just increasing batch size. 🧵⬇️

(2/2) 🔗 nvda.ws/4mFl7GO (describing the GFMBench‑API framework) 💻 nvda.ws/4t6FtdT (reference implementation) 🔬Example study using GFMBench‑API ⬇️ nvda.ws/4eApelm

(1/2) 🧬 The evaluation of genomic foundation models is scattered - GFMBench‑API brings it together. 🚀 GFMBench-API solves this by providing a universal middleware with a standardized API that unifies evaluation - covering tasks such as regulatory element prediction and variant effect scoring - to enable transparent, scalable comparisons. 🧵👇

NVIDIA CEO Jensen Huang joined the world's leading scientists, technologists, and innovators at the 12th @brkthroughprize ceremony. On the red carpet, he shared why science is a team sport, and what it means to celebrate the people pushing humanity forward. 🔭