generative

MIT’s FrameDiff Generative AI imagines new protein structures that could transform medicine

by
Generative AI imagines new protein structures

The FrameDiff system has been tested on the task of building single proteins and researchers have found that it can create large proteins with up to 500 parts. Unlike previous methods, it does not need to rely on a pre-existing map of protein structure. Credit: Alex Shipps/MIT CSAIL via Midjourney

University of Toronto researchers present scGPT: a basic model for single cell biology based on a generative transformer pre-trained in a repository of more than 33 million cells

by
University of Toronto researchers present scGPT: a basic model for single cell biology based on a generative transformer pre-trained in a repository of more than 33 million cells

https://www.biorxiv.org/content/biorxiv/early/2023/07/02/2023.04.30.538439.full.pdf Natural language processing and computer vision are just examples of the fields where pre-trained generative models have been incredibly successful. In particular, a viable strategy for building foundation models is to combine various large-scale datasets with pre-trained transformers. The study investigates the feasibility of basic models for further research in cell biology and genetics … Read more