Generative Artificial Intelligence-Assisted Protein Design Must Consider Repurposing Potential

Abstract

Generative artificial intelligence software used for chemical and protein design has repurposing potential. We propose careful discussion in the biotech community on security considerations of such technologies and serious consideration of restrictions to control who can access the software and what applications it is used for.

FIG. 1.

Structure of ricin, solved using X-ray crystallography (PDB 3RTJ), computationally refined using PDB-REDO. Chain A is shown in blue with a dinucleotide in the enzymatically active site. Chain B is depicted in green. Illustrated using Chimera1.17.

FIG. 2.

Structure of yeast hexokinase (PDB 1IG8) and an inpainted hallucinated model (backbone generation) was done with RFdiffusion. Primary sequence was generated using ProteinMPNN and the structure prediction was by using AlphaFold. All this was done on Google Colab. An experimentally solved structure of yeast hexokinase is depicted on the left. Residues involved in binding of glucose and ADP (and flanking regions) have not been modified, whereas the rest of the protein has been hallucinated resulting in the model on the right. Whether or not this protein would fold correctly or function as an enzyme is unknown.