Statistical learning approaches for obtaining interpretable reduced representations of multimodal sequencing datasets