Speaker
Description
Peptides are short biomolecules with numerous desirable properties for biomaterials design including multifunctionality and biocompatibility. Over the past decade, there has been an explosion in the use of generative deep learning models for design of general de novo molecular design, including peptides; however, analysis of generative models and design spaces remains an open area of research. In this talk, I will present our lab's ongoing work on the creation and assessment of latent space-associated deep learning models and optimization procedures for peptide design. I will discuss model design and optimization in the context of latent space-associated models, including data curation and semi-supervised learning in the context of designing antimicrobial peptides, for which there is plentiful physico-chemical data but sparse highly-relevant experimental data. I will also discuss model quality analysis, focusing on latent space organization, information content, and visualization, and performance and visualization of optimization procedures within different latent spaces. Finally, I will present preliminary work on bringing together deep learning, molecular dynamics simulation, and experiment to design peptides aggregating into functional amyloid structures. Overall, our work lays necessary groundwork for peptide design procedures and for assessing the quality of generative models.
| Keyword-1 | deep learning |
|---|---|
| Keyword-2 | peptide design |