From a hard question to a dose, with AI agents

An mRNA vaccine QSP project, shown end to end. Six AI agents handle the routine work, and the researcher makes every decision.

Research question How should an mRNA vaccine be dosed and spaced to keep antibodies protective? LNP delivery B-cell response Antigen expression Antibody decay Boosters 1,000 virtual people Evidence screening 1,860 papers found 240 studies kept 1,120 parameter values Each value is traced back to its source paper. Immune network model Injected dose LNP uptake Antigen T helper B cells Plasma cells Antibodies Antibody titre over time Days after the first dose Booster Virtual people Stay protectedFall below the level Months after the second dose Protective level Report draft Methods Calibration Predictions Risks and limits Work plan Frame the question and planCoordinator Review the evidenceLiterature agent Build the immune modelModel agent Calibrate on trial dataCalibration agent Simulate virtual peopleSimulation agent Draft the reportReport agent Extracted parameters LNP uptake rate0.35 per hour mRNA half-life in cells9 hours Germinal centre growth0.5 per day Memory B-cell decay0.01 per day Antibody half-life28 days Model size 38species 54parameters 16equations dM/dt = kin · LNP − kd · M dB/dt = kb · A · Th · (1 − B/Bmax) − kdB · B dAb/dt = kp · Plasma − kab · Ab Parameter uncertainty mRNA half-life Antigen expression B-cell growth Antibody half-life 240 fits ran in parallel. Grey is the range before fitting, blue is after. Protection at 6 months Recommendation Recommended regimen 50 µg, 4 weeks apart Two doses, plus a booster at month 6 if over 75 Predicted 74% protected at 6 months, against 38% at 10 µg Researcher approved ? Researcher

Illustrative example with simulated data. It does not show a real drug or clinical result.