Translation of cellular protein localization by generative adversarial network

Abstract

The protein localization in cells had been analyzed by the fluorescent labeling by indirect immunofluorescence and fluorescent protein tagging. However, the relationships between the localizations between different proteins had not been analyzed by artificial intelligence. In this study, we applied the generative adversarial network (GAN) to generate the protein localizations each other, in which the generation was dependent on the types of cells and the relationships between the proteins. Lamellipodia are one of the actin-dependent subcellular structures involved in cell migration and are mainly generated by the Wiskott-Aldrich syndrome protein (WASP)-family verprolin homologous protein 2 (WAVE2) and the membrane remodeling I-BAR domain protein IRSp53. Focal adhesions are another actin-based structure that contains vinculin protein and are essential for cell migration. In contrast, microtubules are not thought to be directly related to actin filaments. The GAN was trained using images of actin filaments paired with WAVE2, vinculin, IRSp53, and microtubules. Then, the generated images of WAVE2, vinculin, and IRSp53 by the GAN showed high similarity to the real images of WAVE2, vinculin, and IRSp53, respectively. However, the microtubule images generated from actin filament images were inferior, corroborating that the microscopic images of actin filaments provide more information about actin-related protein localization. Collectively, this study suggests that the image translation by the GAN can predict the localization of functionally related proteins.