Computational Protein Design

Computational Protein Design sits within structural biology and addresses de novo and inverse-folding-based design of proteins for desired structure and function. The page below sketches the conceptual scope of the area, the methodological tools it relies on, and the recent literature anchoring its current frontier.

The area organises around a small number of recurring axes: scope (what biological scales the work spans), method (the dominant experimental or computational tools), data regime (what kinds of measurements are now routine vs. still frontier), and open questions (what the field cannot yet do reliably). The sources below cover different combinations of these axes.

Frontier results

A primary recent reference for this area is De novo design of protein structure and function with RFdiffusion (Watson et al., 2023), which contributes to the methodological or empirical conversation that defines the current frontier of computational protein design. It illustrates the kind of question the field is actively pursuing — the specific technical claim, the dataset or system on which it was validated, and the way subsequent work builds on it.

A primary recent reference for this area is Accurate prediction of protein structures and interactions using a three-track neural network (Baek et al., 2021), which contributes to the methodological or empirical conversation that defines the current frontier of computational protein design. It illustrates the kind of question the field is actively pursuing — the specific technical claim, the dataset or system on which it was validated, and the way subsequent work builds on it.

Open questions

Open questions in computational protein design cluster around scaling current methods to larger systems, integrating measurements across modalities, and producing predictive rather than descriptive models. The references above mark the work that the next iteration of this page should engage with in more specific detail.