#18 · Single-cell

Explicit objectives now guide single-cell lineage inference and clustering resolution

The three items share a design principle: each method names its optimization target directly instead of relying on implicit heuristics. The preprint models each cell type as an affine subspace defined by linear transcriptome constraints, a geometric account of cell identity benchmarked on simulated data. Carta infers optimal cell differentiation maps from single-cell lineage tracing data by balancing map complexity against unobserved transitions. CANTAO uses average overlap to compare ranked differentially expressed gene lists, using that metric to guide clustering resolution and annotation in scRNA-seq. What is new is less any single algorithm than the choice of criterion as the central object: a geometric constraint, a complexity penalty, or a rank-overlap metric.