Embeddings

• Extracting protein and genomic embeddings and inspecting what they encode
• Downstream heads for classification, regression and annotation
• Baseline comparison against BLAST and classical features before claiming a gain

Agents

• Agent loops: planning, tool invocation and observation
• Wrapping bioinformatics tools so a model can call them safely
• Error propagation, and why a long autonomous chain compounds mistakes

Pipelines

• Literature retrieval, data acquisition and analysis chained end to end
• Human checkpoints at the steps where an error would be expensive
• Logging every call so an agent-produced result can be audited

Judgement

• Confident wrong answers in biology and the domain checks that catch them
• Cost control and runaway loops in autonomous execution
• Biosecurity considerations in sequence design and the responsible disclosure norm