DeePlexing – Extending Imaging Multiplexity Using Machine Learning

Spatial proteomics and transcriptomics are fast-emerging fields with the potential to revolutionize various branches of biology. In the last five years, various multiplex immunofluorescence and immunohistochemistry imaging methods have been developed to stain 5-60 different protein markers in a given tissue. Nonetheless, most of these techniques are iterative and can image a maximum of 3-8 markers in a single cycle, resulting in processing time of several hours to days.

Scientists at National Institute of Allergy and Infectious Diseases (NIAID) and National Cancer Institute (NCI) have developed a new method - DeePlexing - that uses a deep learning algorithm to dramatically enhance the number of markers stained in a single imaging cycle. This is accomplished with no changes or upgrades to the imaging platform itself. In the DeePlexing method, multiple antibodies/probes are conjugated to the same fluorophores and later deconvolved computationally to retrieve the multichannel signal with high accuracy. In digital spatial profiling, DeePlexing enables the user to detect seven different protein markers in a single cycle using only three fluorophores and even quadruple the number of markers in a single round without compromising the quality of RNA and protein in the sample. For multiplex protein profiling, DeePlexing can potentially stain for up to 255 different protein markets with eight fluorophores and deconvolve the signal for each of the 255 markers computationally.