eduzhai > Applied Sciences > Engineering >

Improved Conditional Flow Models for Molecule to Image Synthesis

  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: In this paper, we aim to synthesize cell microscopy images under differentmolecular interventions, motivated by practical applications to drugdevelopment. Building on the recent success of graph neural networks forlearning molecular embeddings and flow-based models for image generation, wepropose Mol2Image: a flow-based generative model for molecule to cell imagesynthesis. To generate cell features at different resolutions and scale tohigh-resolution images, we develop a novel multi-scale flow architecture basedon a Haar wavelet image pyramid. To maximize the mutual information between thegenerated images and the molecular interventions, we devise a training strategybased on contrastive learning. To evaluate our model, we propose a new set ofmetrics for biological image generation that are robust, interpretable, andrelevant to practitioners. We show quantitatively that our method learns ameaningful embedding of the molecular intervention, which is translated into animage representation reflecting the biological effects of the intervention.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×