A diffusion model conditioned on compound bioactivity profiles for predicting high-content images

High-content imaging (HCI) provides a rich snapshot of compound-induced phenotypic outcomes that augment our understanding of compound mechanisms in cellular systems. Generative imaging models for HCI provide a route towards anticipating the phenotypic outcomes of chemical perturbations in silico at unprecedented scale and speed. Here, we developed Profile-Diffusion (pDIFF), a generative method leveraging a latent diffusion model conditioned on in silico bioactivity profiles to predict high-content images displaying the cellular outcomes induced by compound treatment. We trained and evaluated a pDIFF model using high-content images from a Cell Painting assay profiling 3750 molecules with corresponding in silico bioactivity profiles. Using a realistic held-out set, we demonstrate that pDIFF provides improved predictions of phenotypic responses of compounds with low chemical similarity to compounds in the training set compared to generative models trained on chemical fingerprints only. In a virtual hit expansion scenario, pDIFF yielded significantly improved expansion outcomes, thus showcasing the potential of the methodology to speed up and improve the search for novel phenotypically active molecules.