1/ At Valence Labs, @RecursionPharma's AI research engine, we’re focused on advancing drug discovery outcomes through cutting-edge computational methods Today, we're excited to share our vision for building virtual cells, guided by the predict-explain-discover framework 🧵

In my experience, the biggest risk with interdisciplinary AI projects is solving the *wrong problem*. In my latest piece for Inside @valence_ai I explain how I think AI experts should take more ownership of problem formalization to maximize the chance of success [1/2]

🧬 Bridging the gap in AI drug discovery. @AlisandraDenton, Staff Machine Learning Scientist at Recursion and one of the authors on our recent paper in @NatureBiotech, explains how the AI model TxPert predicts how a cell will respond to perturbations. Predicting a cell’s RNA activity, or transcriptome, is key to bridging the gap between cellular changes and clinical outcomes and advancing the potential for AI drug discovery. As Ali says, “with hundreds of cell types and so much disease variation, the total possibilities are too vast to measure in a lab.” She describes how TxPert allows us to perform a “Virtual Assay,” taking the mathematical signature of a healthy cell called the Basal State and adding the perturbation’s embedding to deliver a highly accurate prediction of what the cell’s transcriptome will look like after treatment. TxPert uses layered graph-based models that integrate phenomics — or how a cell looks — and transcriptomics — which genes are expressed — along with massive public biological knowledge resources. The model can even predict how a perturbation will work in entirely new cell lines it hasn’t seen before as well as accurately forecast the effects of “double perturbations,” consistently identifying "unknown unknowns" that traditional models — and even massive general-purpose AI — often miss. Ali notes that TxPert is currently predicting genetic perturbations, but more flexible models —  including those predicting drug effects — are in the works. 👉 Check out the full paper in Nature Biotech: nature.com/articles/s41587-0…

TxPert: using multiple knowledge graphs for prediction of transcriptomic perturbation effects go.nature.com/3QZW9WQ

🧬 Closing the translation gap between cells and patients. 😷 @NatureBiotech just published a new paper from Recursion on TxPert – a deep learning framework that accurately simulates the transcriptomic shift in unseen biological contexts. TxPert represents an important step in our ongoing work to accurately model transcriptomics and bridge the gap between in vitro discovery and clinical reality – which is critical for improving and scaling AI drug discovery. 🔹 TxPert address this translational gap through: ▪️ Graph Neural Networks (GNNs): Rather than treating genes as isolated lines of code, TxPert uses an advanced Exphormer-MG architecture to map genetic perturbations across multiple, massive knowledge graphs, forcing the model to understand both the physical (phenomic) and molecular (transcriptomic) realities of a cell simultaneously. ▪️ Simulating the "latent shift": By mathematically applying a "perturbation embedding" to a cell's baseline state, TxPert can accurately predict the entire post-perturbation transcriptomic profile—without anyone ever having to touch a pipette. ▪️ Predicting unseen biology: TxPert successfully predicts the transcriptomic outcomes of completely unseen single perturbations, complex combinatorial therapies (Double Perturbations), and even how known drugs will act in entirely new, unseen cell lines. TxPert is one of several models at Recursion to model transcriptomics and close the translation gap between cell responses and patients in the clinic. 🎉 Congrats to the team! Frederik Wenkel, Wilson Tu, Cassandra Masschelein, Hamed Shirzad, Liam Hodgson, Ihab Bendidi, Cian Eastwood, Shawn Whitfield, Craig T. Russell, Yassir El Mesbahi, Marta Fay, @bertonearnshaw, @ENoutahi, and @AlisandraDenton. 👉 Read the full publication in Nature Biotech here: nature.com/articles/s41587-0…

We had a fantastic time with @valence_ai at @iclr_conf in Rio sharing our latest machine learning breakthroughs, including presentations on TxFM, our state-of-the-art transcriptomics model that outperforms models up to 100x larger in terms of data size, and MarS-FM, our new class of generative models for molecular dynamics simulations. And there were lots of great community conversations happening at the rooftop TechBio Social, co-hosted with ICLR’s Learning Meaningful Representations of Life (LMRL) Workshop. Coming soon: we’re looking forward to sharing more of our ML breakthroughs at @icmlconf! 👉 TxFM paper here: openreview.net/pdf?id=NqZqCl… 👉 MarS-FM paper here: arxiv.org/html/2509.24779v3

illustrious researchers interested in MD ⁦@ylecun⁩