AI and the World's Hard Problems · Life sciences · arXiv · 2026-10-20
scCTM: Continuous Thought Machine Reconstruction of Glioma Cell Fate Dynamics Attractor Landscapes and Algorithmic Perturbation Signatures
Cell fate plasticity in glioma is usually inferred from static single-cell snapshots. The authors built scCTM, which combines the Sakana AI Continuous Thought Machine with autoencoding to predict single-cell fate dynamics, and tested it on two H3K27M malignant cell datasets. The CTM autoencoder reached an AUROC of 0.99 on both, and a perturbation analysis highlighted recurrent high-impact network interactions involving genes such as ARID1B and NRXN1. These are in silico findings.