Figure 2.

The sigmoid filter prevents variations in a single dominant ADT feature from affecting clustering results

We investigated the effectiveness of the sigmoid filter in preventing variations in a few individual dominant ADT features from affecting clustering results using a real CITE-seq dataset (Dataset 1).

(A and B) Clustering using the LinQ-View model with (B) and without (A) the sigmoid filter. Two models achieve identical results on the majority of the dataset except for CD4⁺ T cells. A naive integration without the sigmoid filter failed to distinguish naive CD4⁺ T and memory CD4⁺ cell subsets (indicated by a dashed circle).

(C) RNA modality contribution of distance between memory CD4⁺ T and naive CD4⁺ T cells with and without sigmoid filtering.