We next compared the performance of bridge integration against two recently proposed methods for integrated analysis of multimodal and single-modality datasets. Both multiVI49 and Cobolt50 utilize variational autoencoders for integration, and while they do not explicitly treat multi-omic datasets as a bridge, they aim to integrate datasets across technologies and modalities into a shared space. When applied to the previously described datasets, both methods were broadly successful in integrating scRNA-seq and scATAC-seq data, but did not identify matches at the same level of resolution (for example, neither method successfully matched ASDC in scATAC-seq data to the ASDC in the Azimuth reference) (Fig. 3b and Supplementary Fig. 2d-f). We also found that the latent space and neighbor relationships learned by bridge integration was most consistent with the labels originally assigned in ATAC-seq analysis (Supplementary Fig. 2c). When comparing computational efficiency, bridge integration (0.8 hours, not including 1.2 hours of preprocessing time), and Cobolt (3.3 hours) were the most efficient methods, while multiVI required more computational resources (15.7 hours).
