Fig. 3
From: scFTAT: a novel cell annotation method integrating FFT and transformer
Ablation experiments of scFTAT across six typical scRNA-seq datasets. A, C and E depict the results on three human tissues, B, D, and F for three mouse tissues. “Transformer-LDA” refers to the configuration where LDA is removed from the Transformer module and replaced with traditional HVGs, “Transformer” for keeping LDA to the previous modules, “Transformer + FFT” for adding an FFT module to the Transformer; “scFTAT” includes the modules above, along with kernel approximation, rotated attention matrix, and other fine-tuning modules to optimize the entire model