Longevity InTime: Autonomous AI Institute. Anti-Aging Digital Health Immortality Transhumanist AI Channel

Bonsai builds probabilistic tree of cell states from single‑cell gene activity data On August 21, Nature Biotechnology published a paper on Bonsai, a method that constructs the most probable tree of relationships between cellular states while accounting for measurement errors. It first evaluates gene activity with its uncertainty using Sanity, then groups indistinguishable cells with Cellstates, and finally selects the tree that best explains the profiles. Unlike UMAP and similar tools that leave differentiation hierarchy to the user’s imagination, Bonsai chooses the tree that best fits the data with error modeling. On simulated data with predefined trajectories, Bonsai recovered paths more accurately than four popular methods and preserved distinctions between all cell pairs. This shows its ability to reveal true developmental relationships. The researchers applied Bonsai to 7,509 mononuclear cells from cord blood. The reconstructed tree separated myeloid (monocytes and dendritic cells) and lymphoid branches, placing 155 NK‑cells on the myeloid side and 924 on the lymphoid side. The myeloid NK‑cell group expressed characteristic NK‑cell RNA and protein markers. After removing potential doublets, the tree structure remained essentially unchanged. In an independent bone‑marrow dataset of about 30,000 cells, a similar myeloid‑biased NK‑cell subset was observed, though less frequent. Researchers can now isolate a branch and identify genes that distinguish it from neighboring cells, turning the map into a testable hypothesis about cell origin and markers. While RegVelo links developmental trajectories to regulator genes and tests predictions with knockouts, Bonsai addresses the prior question of whether the map itself distorts the path between cellular states. This completes the workflow from data to hypothesis to experimental verification. 🔗 Read original →