Post 41: The first application of neural networks with proteins 📄

Although it might be thought that the use of artificial intelligence applied to protein science is a novel idea, in reality, the first works in this field can be traced back ~30 years ago with works like that of Burkhard and Chris, who in 1992 trained a neural network to predict if an amino acid corresponded to a region of secondary structure (i.e., alpha helix, beta sheet, or turn), achieving results comparable to an experimental technique known as circular dichroism. What a stir it must have caused back then? 🤔

30 years later we can do a lot of things with AI, from predicting 3D structure to designing biological functions. Where will we be in another 30 years?

Even in 1982, neural networks were already used to predict protein translation sites. It must have been quite complicated to train models like this with the hardware of the time.

Refs:

1. Prediction of protein secondary structure at better than 70% accuracy
2. Use of the ‘Perceptron’ algorithm to distinguish translational initiation sites in E. coli