ER-WGAN: Prediction of Cell Painted Endoplasmic Reticulum from Brightfield Images

Abstract Prediction of cell painting facilitates highthroughput screening of cellular biology and disease mechanisms, accelerating the discovery of novel therapeutic targets or drugs. This study explores the prediction of cell painted Endoplasmic Reticulum (ER) images from transmitted light brightfield images employing deep learning techniques. A conditional Generative Adversarial Network (cGAN) based framework incorporating Wasserstein loss and Gradient Penalty with a modified UNet++ based generator and a patch discriminator is used to predict cell-painting images from brightfield images captured at varying focal planes. Evaluation of the GAN network reveals promising results with a Mean Absolute Error (MAE) of 0.20, Multi-Scale Structural Similarity Index (MS-SSIM) of 0.80, and Peak Signal-to- Noise Ratio (PSNR) of 56.52. The results showcase the effectiveness of the proposed approach in accurately predicting the ER cell painting channel from transmitted light microscopy. Consequently, the study proposes the ER-WGAN network for predicting cell painting using brightfield images.