Conventional Protein engineering vs Computational Engineering of Proteins

Source : https://www.amgen.com/stories/2019/09/designing-proteins-from-scratch

Most protein design pipelines are not trying to design a binder or a protein with specific function in one shot. Computational protein design is tightly coupled with experimental screens. A protein design pipeline can include many different neural networks combined together or neural networks combined with traditional modeling tools. With available models and tools, the pipeline for computational protein design differs from one family of proteins to other.

AlphaFold reveals the structure of the protein universe

Language models of protein sequences at the scale of evolution enable accurate structure prediction

Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, Alexander Rives

De novo protein design by deep network hallucination

I Anishchenko, TM Chidyausiku, S Ovchinnikov, SJ Pellock, D Baker. De novo protein design by deep network hallucination. (2020) bioRxiv, doi:10.1101/2020.07.22.211482.

Source: https://www.biorxiv.org/content/10.1101/2020.07.22.211482v1.full.pdf

Protein sequence design by explicit energy landscape optimization

C Norn, B Wicky, D Juergens, S Liu, D Kim, B Koepnick, I Anishchenko, Foldit Players, D Baker, S Ovchinnikov. Protein sequence design by explicit energy landscape optimization. (2020) bioRxiv, doi:10.1101/2020.07.23.218917.

Using the above method we can derive a protein sequence that will fold into a desired shape. But that is not always a guarantee of function. If a protein was designed to have affinity for a receptor, we would want to computationally predict which one of the sequences is a better fit for a receptor.

Designing with TERMs

https://www.mlsb.io/papers_2021/MLSB2021_TERMinator:_A_Neural_Framework.pdf