Small But Mighty: What an LSTM Network Reveals About Protein Design and About Machine Learning 

Presented by Joanna Slusky on September 11th 2024
abstract: Long Short Term Memory (LSTM) recurrent models are humble and relatively small. We tried to use an LSTM to generate novel proteins, focussing on the transmembrane beta barrel fold. It went a lot better than you would think! Also, the small size of the model allowed us to vary parameters to determine what controls different parts of protein sequence learning.