Do deep generative models in single-cell omics really work for perturbation prediction?. Some benchmark studies say yes:.🔗 🔗Others say no:.🔗 🔗 BMC Genomics: To move beyond this

Really liked this meme from @lucascamillomd 😅

🚀 What do genomic Transformers actually learn about biology?.•What knowledge do they hold at random init, after pre‑training, and following fine‑tuning?.•We dove deep into every attention head to find out. 📄 Preprint live now “Interpreting Attention Mechanisms in Genomic

Exciting day for the virtual cell revolution! 🚀 . @arcinstitute has just launched the Virtual Cell Challenge, calling on leading institutions worldwide to push the boundaries in virtual cell modeling (. Huge congrats to Arc Team (@genophoria @pdhsu.

What a mind-blowing week for AI in biology!. 🚀 Xaira just dropped the largest genome-wide perturbation dataset ever last week. 🧬 Arc unveiled STATE yesterday, the most powerful foundation model for single-cell perturbation prediction. 🧠 DeepMind followed up with AlphaGenome,.

More in-depth technical tweetorial from the amazing @_annhuang , the first author of our X-Atlas/Orion paper! .

My take?. It’s still early for AI models to accurately simulate biological processes. The bottleneck: lack of specialized, dynamic training data. But the potential is enormous—for science education, interactive textbooks, or even virtual cell simulation models.

For comparison, here’s one of the best educational demos of the central dogma I’ve seen—made by humans, of course:. 🔗

Next, I tried the latest @midjourney model, which supports video generation up to 5 seconds. Here’s what it came up with for the same prompt. 🎥⬇️

First up: Google’s Veo 3 — arguably the most advanced video model right now. It supports up to 8 seconds of video generation. Here’s what Veo gave me for “DNA to RNA to Protein.”.🎥⬇️

Prompt I used: "Generate an high-resolution animation video showing the central dogma of cellular biology, specifically how a DNA sequence is transcribed into RNA sequence, and how the RNA sequence is translated into protein sequence in a cell.".

AI models for video generation are advancing fast. But how well can they capture biological processes?. I put them to the test using the most iconic concept in molecular biology:.The central dogma — DNA → RNA → Protein. Let’s see what the models produced 🧵

Highly recommend! I’ve had the privilege of working with both @sara_mostafavi and @anshulkundaje when I was a student—they’re true pioneers in the field.

If you're passionate about building the world's first AI virtual cell model, we're hiring at @Xaira_Thera-- come join us!.

RT @DeryaTR_: This is an incredible public dataset, 8.4 million single cells with 20,000 protein perturbations! Truly a game-changer for Bi….

Got an email from my PhD student today—she had traveled to Tehran just a few weeks ago, right before war was unfolding. No complaints. No requests. Just a quiet, steady project update—sent with the kind of calm strength that left me speechless. I sat there staring at the.

🚀 Xaira Therapeutics has just dropped a game-changer for AI-driven biology. Today, we unveiled X-Atlas/Orion, the largest publicly available genome-wide Perturb-seq dataset to date—spanning 8.4 million single cells with perturbations across all ~20,000 human protein-coding

🏆 Congratulations to nnInteractive (from German Cancer Research Center) and BioMedparse (from @Microsoft ) for taking the top prize in the CVPR 2025 Foundation Models for Interactive and text-guided 3D Biomedical Image Segmentation Challenge, respectively!. We curated 286K for

Woke up to this kind message in my inbox and it truly made my day 🙏🥹. As researchers, we often work long hours not knowing who might read or benefit from our papers. Seeing scGPT inspire others to build new models and push the field forward is one of the most meaningful