ProteoGAN

ProteoGAN employs conditional generative adversarial networks to generate protein sequences conditioned on hierarchical Gene Ontology functional labels for targeted protein design and exploration of functional sequence space.

Key Features:

• Conditional Generative Modeling: ProteoGAN uses conditional generative adversarial networks to generate protein sequences based on specified functional labels.
• Hierarchical Gene Ontology Integration: The model conditions generation on hierarchical functional labels from the Gene Ontology to capture relationships among biological functions.
• Advanced Evaluation Metrics: The developers introduced a suite of biologically and statistically inspired metrics to assess generative model performance.
• Model Insights and Ablation Studies: The framework includes analysis and ablation studies to evaluate the influence of hyperparameters on model performance.
• Potential for Novel Protein Functions: Combining different functional labels permits exploration and generation of protein sequences with potentially novel functions.

Scientific Applications:

• Therapeutic development: Generate protein sequences with specified functions to support the development of novel therapeutics.
• Enzyme engineering: Design enzyme sequences annotated for desired catalytic functions for biotechnology applications.
• Biomaterials development: Generate protein sequences with targeted functions for biomaterials applications.
• Protein structure–function studies: Explore sequence–function relationships for fundamental research into protein biology.
• Synthetic biology: Propose candidate protein sequences conditioned on functional labels for synthetic biology research.

Methodology:

Train a conditional generative adversarial network on datasets of existing protein sequences conditioned on Gene Ontology functional labels (including hierarchical labels); evaluate outputs using biologically and statistically inspired metrics and perform ablation studies to assess hyperparameter effects.