Figure 5.

Effects of altering individual parameters on cluster annotation algorithm. Plots shown are the cumulative distribution functions of the AUC_{Ranks 1–5} metric. Each panel illustrates the effect of modifying a single parameter on the algorithm performance. (A) Modifying the source of reference cells used to compute CDGs for downstream analysis has a limited impact on performance, with the pan-study and within study options slightly outperforming the within tissue option. (B) Modifying the number of CDGs considered has a strong impact on performance, with fewer genes (e.g., 1, 3, or 5) showing considerably worse performance. (C) Modifying the local score version used (absolute or scaled) has a mixed effect, with scaled GCAs contributing more of the worst-performing (left tail) and best-performing (right tailed) parameter combinations. Note that only the parameter combinations which used an L2 norm-based rank or the composite rank were included in this analysis, as only absolute GCAs were considered for the modified L0 norm-based ranking. (D) Modifying the weighting method for calculating the L2 norm has minimal impact on performance, with weighting by either fold change (FC) or log₂FC slightly outperforming the unweighted method. (E) Modifying the final ranking method has a strong impact on performance, with the modified L0 rank and L2-based rank showing the worst and best performances, respectively.