i like science and cats. bioinformatics @xaira_thera
Joined September 2020
- Tweets52
- Following107
- Followers149
- Likes106
Pinned Tweet
1/12
Excited to share our team's latest work and the first @xaira_thera preprint! Here, we introduce FiCS Perturb-seq, an industrialized platform for generating scaleable, high-quality perturbation data.
📄 Read the preprint: biorxiv.org/content/10.1101/…
Ann Huang retweeted
Meet X-Scientist: our step toward a self-improving virtual drug hunter!
Today, we’re introducing Xaira’s agentic AI platform for drug discovery:
🧠 Beats Codex using the same Astra model by 7.4 percentage points on DrugDiscoveryBench.
🧬 Nominates targets ~10× faster than human curation, with a comparable wet-lab hit rate.
🔁 Conducts autonomous research, outperforming AutoResearch and AutoScientists in the PerturbQA benchmark.
Here’s what we built, what we tested, and where we’re going. 🧵
Ann Huang retweeted
Two preprints. Two days apart. Same mission: crack the genotype × transcriptome code at single-cell resolution.
Let's break this down:
Saurty-Seerunghen, Lee et al. (Nam, Martelotto & Grillo teams) – GoT-Multi-Gap for RNA-templated gap filling. Clever enzymatic trick: BstFL polymerase does RT + nick translation to gap-fill between flanking probes on RNA, then ligates. Bolts directly onto probe-based scRNA-seq (think 10x Flex-type workflows). Whole transcriptome + mutation calls from the SAME cell. Multiplexed — target many mutations at once Leverages existing probe-based platforms → low barrier to adoption
biorxiv.org/content/10.64898…
Blattman, Maslah et al. (MSK/Pe'er, Charlinge & Lareau teams) – GIFT Genotyping In Fixed Transcriptomes. This one is a biobank's dream. >99% genotyping accuracy — that's wild. Scales to HUNDREDS of mutations per cell. and THE best part...it works on FFPE 🤯 — unlocks archival clinical samples. Clonal lineage tracing in heterogeneous tumors.
biorxiv.org/content/10.64898…
So how do they compare? Gap filling → elegant chemistry, seamless integration with fresh-sample probe-based scRNA-seq workflows. Minimal disruption to existing pipelines. GIFT → sheer scale + FFPE compatibility. If you have a tumor biobank, this is your ticket to retrospective clonal mapping.
Both obliterate the old limitation of choosing between genotype OR transcriptome. The convergence of these approaches means we're entering an era where every single-cell experiment can also be a clonal evolution experiment. 🚀
Ann Huang retweeted
1/7 First of all, big shoutout to co-authors on modeling (@MKarimzade, @neal_ravindra, @RexMa9, @HAOTIANCUI1, @LeeTaliq), huge appreciation to data generation (Lexi, @alerasool, Adam) and bioinformatics team (@_annhuang), and leadership for vision and direction (@BoWang87, @inCiChu)!
Preprint is now live on bioRxiv: biorxiv.org/content/10.64898…
All models start from high-quality data.
Our X-cell is up at @biorxiv_bioinfo !
Read our full paper at biorxiv.org/content/10.64898…
Part of the data and the model weights will be shared soon. stay tuned!
Ann Huang retweeted
Massive push from the dream team 🫡 walkthrough coming soon!
Today we’re announcing X-Cell — Xaira’s first step toward a virtual cell. 🧬
A foundation model that predicts how gene expression changes under causal perturbations — across cell types, conditions, and even unseen biology.
This is not trained on observational atlases.
It is trained on interventions.
🧵👇
Ann Huang retweeted
1/ So excited to have had the opportunity of contributing to this magnificent effort! Foundation models of observational transcriptome often memorize gene co-expression networks without understanding the underlying logic. Genetic perturbation datasets make it possible to
Today we’re announcing X-Cell — Xaira’s first step toward a virtual cell. 🧬
A foundation model that predicts how gene expression changes under causal perturbations — across cell types, conditions, and even unseen biology.
This is not trained on observational atlases.
It is trained on interventions.
🧵👇
Ann Huang retweeted
Most AI in biology is trained to observe. At @Xaira_Thera, we’re training it to intervene.
In collaboration with @BoWang87 and team, we’re proud to launch X-Cell, a first-of-its-kind virtual cell model.
The breakdown:
🧬 Data: 25.6M cells across 7 genome-scale Perturb-seq campaigns, 16 diverse biological contexts (via X-Atlas/Pisces).
⚡ Model: 4.9 billion parameter diffusion language model with multimodal prior knowledge.
🎯 Performance: 5x performance leap over SOTA on key predictive metrics. Zero-shot generalization to human T cells and unseen cell types.
Predicting unseen biological experiments is a massive step toward true in silico drug discovery. Huge congrats to the whole team! 🚀
Ann Huang retweeted
Today we’re announcing X-Cell — Xaira’s first step toward a virtual cell. 🧬
A foundation model that predicts how gene expression changes under causal perturbations — across cell types, conditions, and even unseen biology.
This is not trained on observational atlases.
It is trained on interventions.
🧵👇
Ann Huang retweeted
In 2022, we ran the largest competition in single-cell genomics on predicting modality profiles across time, hoping that this task provides insight into gene regulation. 27,000+ submissions and 3 years of analysis later, we finally have some insights: bit.ly/4bcZcm8
Virtual Cell community - this one's for you! X-Atlas/Orion is now live on Hugging Face.
Train your own models with streamlined workflows built into the Hugging Face API.
🔗 HuggingFace: huggingface.co/datasets/Xair…
📜 License: cc-by-nc-sa-4.0
Ann Huang retweeted
New to virtual cells? @freethink’s primer explains these AI-powered, in silico models—why they’re game-changers for biology & the hurdles still to solve. Worth a read bit.ly/4ljre1j
In the wake of all the Bio x AI updates, I'm also here to spread awareness of important cat-related research.
Isparta et al analyzed 400+ cat videos and found that cats favor the leftward side (p < 0.001), allowing for fast response to stimuli.
Paper: cell.com/current-biology/ful…
Ann Huang retweeted
We are looking for talented postdocs/visiting students to join our team @MITdeptofBE and @kochinstitute to innovate spatial multi-omics technologies, map clinical samples, and integrate AI/ML models to advance next-generation cancer immunotherapy. Please help to spread!!
Such an exciting initiative. Congrats to all those involved!
Hope people put the X-Atlas/Orion dataset to use 💪
Register today for the Virtual Cell Challenge and use AI to solve one of biology’s most complex problems.
Announced in @CellCellPress, the competition is hosted by Arc Institute and sponsored by @nvidia, @10xGenomics, and @UltimaGenomics.
1
7
404
11/12
In conclusion, FiCS Perturb-seq is a robust platform for executing large-scale perturbation screens, enabling the creation of datasets like X-Atlas/Orion to accelerate causal foundation models in predictive biology.
📊 Download X-Atlas/Orion: doi.org/10.25452/figshare.pl…
12/12
This work is the result of an incredible collaboration between @Xaira_Thera and @ForesiteLabs! Many thanks to all the amazing co-authors (see preprint for full list): @Stanley_TH @jiangzhuzime @qtaznangel @awblocker and @inCiChu