Seminars

Developing therapeutics and vaccines requires a deep understanding of human biology, often inadequately captured by animal or in vitro models. To address this, we created AutoTransOP, a neural network autoencoder framework that maps omics profiles across species or biological contexts into a shared latent space, identifying relevant molecular features without direct mapping. AutoTransOP leverages artificial intelligence to enhance the translation of experimental findings to human clinical applications.

Directed protein evolution is vital for biomedicine but faces challenges like experimental complexity and inefficient optimization. While protein language models can guide fitness landscapes, they struggle with diverse protein families. We present EVOLVEpro, a few-shot active learning framework combining PLMs and regression models to enhance protein activity. It achieves up to 100-fold property improvements across six proteins in RNA, genome editing, and antibody applications, opening new possibilities for AI-guided protein engineering.