Agentic genomics: Your science, made executable¶
Manuel Corpas, INFLAMomx Training School, University of Salerno, 7 September 2026.
This lecture asks what changes when a scientist can express a biological question in a form that an AI agent can help execute. It uses real RNA-seq and single-cell examples to show how scientific intent becomes an inspectable result, and why clinical-grade agentic genomics requires deterministic execution, traceable evidence, validation and equity.
Watch the full lecture on YouTube
Three ideas to remember¶
- Agentic execution is now a question of when, not if.
- Early adoption gives scientists a chance to shape the field while its methods are still forming.
- The language of agents is context engineering: question, data, design, stopping rules and evidence.
Examples in the lecture¶
- A paired dexamethasone RNA-seq analysis using four donor-derived airway smooth-muscle cell lines.
- A deliberately invalid control that stops without producing a differential-expression result.
- Recovery of the published STMN2 response after TDP-43 depletion.
- A public PBMC3k single-cell workflow, with marker evidence treated as a hypothesis rather than a finished cell identity.
- A benchmark showing that authored scientific rules and deterministic execution contribute different kinds of value.
- An equity question: whose science is missing from the validation?
ClawBio is an open-source scientific commons for inspectable biological AI skills. As checked on 7 September 2026, the public repository had 1,126 stars, 261 forks and 48 commit-contributor accounts.