Fig. 2: CISI-IMC workflow.

a Schematic of the compression and decompression process using CISI-IMC. The barcoding matrix encodes the scheme for barcoding of \(p\) antibodies with \(m\) unique composite channels. By selecting an \(m\) smaller than \(p\), single-cell protein expression data of \(p\) proteins can be compressed into single-cell composite data of \(m\) composite channels. To enable the decompression, single-cell protein expression data of \(p\) proteins is decomposed into a dictionary of \(d\) protein-expression modules and single-cell module activity with sparse activation of the modules for each cell. Decompression is performed by estimating the sparse single-cell module activity using the known barcoding matrix and the dictionary. b CISI-IMC workflow. During the CISI-IMC experimental workflow, a single-cell IMC data from a training dataset and the SMAF algorithm are first used to generate the dictionary. Second, the barcoding matrix is trained on a portion of the training dataset by simulating the single-cell composite data with randomly generated barcoding matrices, and simulated single-cell composite data are then decompressed back to single-cell protein expression data, which are compared to the original single-cell protein expression data. The barcoding matrix with the best decompressing performance is selected for the next steps. Third, antibodies are labeled with the specific combinations of metal isotopes according to the selected barcoding matrix. Finally, tissues of interest are stained with the barcoded antibodies and imaged with IMC. Obtained composite images are segmented into single-cell composite data that are decompressed back to single-cell protein expression data using the pre-trained dictionary.