Reference-guided pseudotime inference across species and biological contexts

Biological age can differ significantly among cells collected at the same chronological age, due to variations in differentiation timing, maturation speed, and degeneration rates. Current pseudotime inference techniques either disregard chronological time data or necessitate accurate time-series labels within comparable species or biological situations.

Consequently, both approaches often struggle to correctly order cells across biological contexts lacking reliable time labels along a relevant axis, such as human embryonic development or disease advancement. In response, researchers introduce Cavebear, a machine learning framework that utilizes scRNA-seq time-series profiles from a reference species or condition to guide pseudotime inference in a target species or condition.

Cavebear surpasses existing methods in accuracy for developmental pseudotime inference and enables in vivo temporal mapping for in vitro experiments. Importantly, the study demonstrates Cavebear's potential in examining cellular-level disease progression in human patients by employing mouse cancer development models as references.

By transferring temporal information across species and conditions, Cavebear facilitates systematic exploration of biological variation in contexts where such annotations have been previously unobtainable.