Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 10·5 min read
Genotypic Triggers: Exposing Pharmacogenomic Blind Spots via Host-Specific Backdoors in Generative Antimicrobial Peptide Models
TL;DR
Genotypic Triggers expose vulnerabilities in antimicrobial peptide models by introducing a backdoor that increases immunogenicity risk for specific genetic profiles. This manipulation raises risk scores by 743% for targeted individuals while maintaining desirable peptide properties.
✦ Why It Matters
Review your peptide generation models for potential genetic vulnerabilities to ensure safety across diverse populations.
Key Takeaways
- Incorporate genetic profiling in peptide validation processes.
- Assess existing models for susceptibility to backdoor attacks.
- Enhance safety protocols to account for genetic diversity.
- Monitor immunogenicity risk scores in drug design.
How It Works
The Genotypic Trigger operates by intentionally altering the generative distribution of peptide models. By embedding specific genetic markers into the training data, the model is manipulated to favor peptides that are predicted to provoke stronger immune responses in individuals with certain HLA alleles.
This targeted approach allows for the identification of peptides that may be safe for the general population but pose risks for genetically predisposed individuals.