Fig. 5

Ranked integration method performance across prediction tasks. Integration methods were ranked by averaging their overall performance across datasets for each prediction task (trajectory inference with PAGA: green, perturbation classification: blue, and classification of disease status: pink). Ranked scores were computed for several metrics for evaluating a prediction task: \(\mathrm {TI}_{\text {corr}}\), F1 score, balanced accuracy (\(\mathrm {acc}_{\text {b}}\)), and area under the receiver operator curve (AUC). Here, higher ranked method scores are indicated by a longer lighter bar. A Overall quality of spliced and unspliced integration performance according to several metrics for evaluating prediction tasks. B Overall quality of moments of spliced and RNA velocity integration performance according to several metrics for evaluating prediction tasks. Of note, CellRank was not performed on unspliced and spliced integration, as it relies on RNA velocity data. Across all three prediction tasks, unspliced integration outperforms unintegrated data, while RNA velocity integration often achieves increased trajectory inference correlation and perturbation classification scores