How do we know whether a biomedical foundation model has learned something biologically meaningful - rather than simply learned to excel on a narrow benchmark? See our new Perspective by @JulioSaezRod and Philipp Sven Lars Schäfer with Nikolas Kalavros & Gustavo Stolovitzky
This work was led by @psl_schaefer with support from Leoni Zimmermann, P. Leo Burmedi, @tanevski , Miri Adler and her team (Avia Walfisch, Noa Goldenberg, Shira Yonassi, Einat Shaer Tamar). @JulioSaezRod and @roramirezf94 supervised the work.
🧵 New preprint: Logic models can simulate cancer signalling & screen drug targets in silico, but they need perturbation data & hand-curated networks patients lack.
We present FICUS for building dynamic logic models directly from patient omics.
📄 doi.org/10.64898/2026.06.19.…
FICUS bundles COSMOS + CellNOpt into one modular, data-driven workflow for patient stratification & mechanistic hypothesis generation in precision oncology.
📦 R package: github.com/saezlab/ficus