SCREENFunction-based synthesis screening
Developing, testing and deploying function-based screening for nucleic acid and protein synthesis. Measuring and adapting AI model capabilities for predicting molecular properties from sequence to identify hazardous functions.
- Evaluating AI protein design methods for generating highly divergent functional protein variants
- Developing methods to map the architecture and components of an unknown sequence
- Identifying functional signatures of hazardous protein families based on conserved molecular features
- Mapping structural neighborhoods of dangerous proteins to calibrate structure-based detection
- Evaluating sequence-to-function prediction on distant AI-designed protein variants
- Identifying novel hazards by predicting protein-protein interactions with known host targets
- Exploring model representations and interpretability for recognizing hazardous functions
Why function-based screening
DNA synthesis screening is central to detecting hazardous biological function before it’s built. DNA sequences encode a highly compressed representation of biological function, which is executed by the cellular environment through the processes of molecular biology. Screening sequences for hazardous functions has traditionally relied on homology to known genes and proteins, but this approach is being strained as synthetic biology pushes beyond the space explored by nature.
Biological foundation models can learn a rich and nuanced representation of biomolecules by mapping their sequence–structure–function relationships to a high-dimensional embedding space. Fourth Eon is evaluating foundation model capabilities for predicting structure, binding, function, and other molecular properties directly from sequence, and exploring how this could be used for detecting biological threats that do not closely resemble anything found in nature.
AI protein design risk
Biological foundation models and biodesign tools are transforming our ability to create novel proteins and pushing the boundaries of synthetic biology. They also pose new challenges for biosecurity due to the potential for engineering dangerous functions.
Fourth Eon is evaluating AI-enabled protein design in the context of biosecurity screening, and seeking to understand the capabilities of current models in generating novel proteins that could challenge existing safeguards.
