In silico spatial transcriptomic editing at single-cell resolution

MotivationGenerative artificial intelligence (AI) has enabled fundamental breakthroughs in visual content creation by text-guided editing. However, the utility of such generative models remains largely understudied for processing increasingly complex bioimage data. ResultsWe propose to algorithmically edit gene expression data to drive cell-level morphological transitions using Generative Adversarial Networks (GAN) and GAN Inversion models. Leveraging cutting-edge spatial transcriptomic datasets with subcellular in-situ resolution and matched high-content imaging data, we propose an in-silico approach to quantify, model and imitate pathological processes in real-life clinical tissue samples. Availability and implementationThe code and video demo is accessible via https://github.com/CTPLab/In-silico-editing