Nicholas Ho

Excited to showcase the science and people behind our GigaTIME paper, recently published in Cell. Paper: aka.ms/gigatime-paper Model: aka.ms/gigatime-model Code: aka.ms/gigatime-code Azure Foundry: labs.ai.azure.com/projects/gigat…

Thrilled to announce alphagenome-pytorch, an accurate, readable, and careful port of AlphaGenome's architecture and weights to PyTorch. Work with @gtcaa @m_kjellberg @chriswzou @tuxinming as part of the GenomicsxAI initiative between @anshulkundaje and @pkoo562 labs.

@garykbrixi @abhinadduri Congrats Garyk!

Evo 2 is out in Nature today, showing that genome language models can predict and design across the full complexity of life, from phages to eukaryotes. A few surprises from the project, including how ignoring trillions of nucleotides was key to getting a good model. 🧵

AlphaGenome is out in @nature today along with model weights! 🧬 📄 Paper: nature.com/articles/s4158… 💻 Weights: github.com/google-deepmin… Getting here wasn’t a straight path. We sat down @googledeepmind to discuss the story behind the model, paper & API: youtu.be/V8lhUqKqzUc

Honored and grateful to be named an ACM Fellow. cs.cmu.edu/news/2026/ma-a…

Super excited to see that @genbioai was highlighted in Dr. @LisaSu's keynote talk. Thank you so much, @AMD — we look forward to what we will achieve together in 2026!

Together with Emma Dann, we are thrilled to present a massive new Perturb-seq atlas of 22M primary CD4+ T cells, from 4 donors, across 3 timepoints – the result of a decade-long collaboration between the Marson (@MarsonLab) and Pritchard (@jkpritch) labs. 🧵👇

MBZUAI’s Institute of Foundation Models has released K2-V2, a 70B reasoning model that is tied for #1 in our Openness Index, and is the first model on our leaderboards from the UAE 📖 Tied leader in Openness: K2-V2 joins OLMo 3 32B Think at the top of the Artificial Analysis Openness Index - our newly released, standardized, independently assessed measure of AI model openness across availability and transparency. MBZUAI went beyond open access and licensing of the model weights - they provide full access to pre- and post-training data. They also publish training methodology and code with a permissive Apache license allowing free use for any purpose. This makes K2-V2 a valuable contribution to the open source community and allows more effective fine-tuning. See links below! 🧠 Strong medium-sized (40-150B) open weights model: At 70B, K2-V2 scores 46 on our Intelligence Index with its High reasoning mode. This puts it above Llama Nemotron Super 49B v1.5 but below Qwen3 Next 80B A3B. The model has a relative strength in instruction following with a score of 60% in IFBench 🇦🇪 First UAE entrant on our leaderboards: In a sea of largely US and Chinese models, K2-V2 stands out as the first representation of the UAE in our leaderboards, and the second entrant from the Middle East after Israel’s AI21 labs. K2-V2 is the first MBZUAI model we have benchmarked, but the lab has previously released models with a particular focus on language representation including Egyptian Arabic and Hindi 📊 Lower reasoning modes reduce token use & hallucination: K2-V2 has 3 reasoning modes, with the High reasoning mode using a substantial ~130M tokens to complete our Intelligence Index. However, the Medium mode reduces token usage by ~6x with only a 6pt drop in our Intelligence Index. Interestingly, lower reasoning modes score better in our knowledge and hallucination index, AA-Omniscience, due to a reduced tendency to hallucinate

Have you heard about the Night Science Podcast, where we talk about the creative process of doing science? We explore this with discussions with brilliant scientists & philosophers and artists, to figure out the tricks of the creative scientific trade. podcasts.apple.com/us/podcast/nig…

I am excited to serve as Proceedings Chair with @kmborgwardt for #ISMB2026 in Washington DC. We have brought together a strong and diverse group of Area Chairs. It is a truly exciting and exhilarating time for #CompBio. Look forward to your paper submissions by the 1/20 deadline

#LLOKI, for cross-platform integration of spatial transcriptomics, is now published in @genomeresearch! 🎉 Work led by @ellie_haber (@mldcmu) and Lane Fellow @SpencerKrieger (@CMUCompBio). Earlier version in #RECOMB2025. LLOKI is pronounced "low-key". genome.cshlp.org/content/early/…

LeJEPA: a novel pretraining paradigm free of the (many) heuristics we relied on (stop-grad, teacher, ...) - 60+ arch., up to 2B params - 10+ datasets - in-domain training (>DINOv3) - corr(train loss, test perf)=95% Paper: arxiv.org/pdf/2511.08544 Code: github.com/rbalestr-lab/l…

🧬 Excited to share Nicheformer out now in Nature Methods! A transformer foundation model linking single-cell & spatial omics, learning spatial context from gene expression to map tissue organization. Led by Ale Tejada & Anna Schaar 👏 👉 nature.com/articles/s4159…

@schwabpa Congrats! Great work!

Discovering interventions that achieve a desired therapeutic effect is a core step in the development of new medicines that remains challenging despite perturbation data being increasingly available in vast quantities. Today, we are announcing the release of the Large Perturbation Model - an in-silico biological discovery model that disentangles perturbations, the contexts in which they were observed, and measurements. Trained entirely on controlled perturbation experiment data, LPM is a generative model of causal effects under intervention that enables researchers to perform a wide range of biological discovery tasks purely in computers, including elucidating the mechanism of action of interventions, their effects across model systems and their similarity to other studied interventions.

Has anyone even asked the cells if they wish to be perturbed

🧬 Building the Virtual Cell starts with data. Today, we’re making the X-Atlas/Orion Perturb-seq dataset even more accessible — now live on Hugging Face @huggingface 🤗 📊 One of the largest & highest-quality perturbation datasets ever released, it provides the foundation for training AI models that can simulate and reason about cellular behavior. 👉 Explore here: huggingface.co/datasets/Xaira… Fewer barriers. More models. A step closer to the virtual cell. 🌌 #VirtualCell #FoundationModels #AIinBiology #HuggingFace

1/3 Metacells boost power in single-cell RNA-seq & multiome analysis. But without checking homogeneity, they risk forming dubious metacells that bias discoveries. We introduce mcRigor: a statistical safeguard for rigorous metacell analysis. 👉 nature.com/articles/s4146…

We are excited to share GPN-Star, a cost-effective, biologically grounded genomic language modeling framework that achieves state-of-the-art performance across a wide range of variant effect prediction tasks relevant to human genetics. biorxiv.org/content/10.110… (1/n)

Welcome to the age of generative genome design! In 1977, Sanger et al. sequenced the first genome—of phage ΦX174. Today, led by @samuelhking, we report the first AI-generated genomes. Using ΦX174 as a template, we made novel, high-fitness phages with genome language models. 🧵