Adversarially Robust Safeguards for Generative Protein Design

AI can now design proteins that preserve the dangerous function of known toxins while looking nothing like them in sequence – and today’s biosecurity screening largely checks sequence similarity. Amirali Aghazadeh from Georgia Tech is developing more robust screening that flags risky designs by their structural and functional similarity, not sequence alone.

Biography

What happens when AI can design a toxin that no screening tool recognizes? Amirali Aghazadeh is an assistant professor of Electrical and Computer Engineering at Georgia Tech, with program faculty roles in Machine Learning, Bioinformatics, and Bioengineering. Trained at Rice with postdocs at Stanford and UC Berkeley, he builds machine-learning methods for protein and small-molecule design — and is now developing biosecurity screening that flags dangerous designs by structure and function, not just sequence.