Efficient generation of hematopoietic progenitor cells from human pluripotent stem cells by robotic automation
Differentiation of human pluripotent stem cells (hPSCs) into therapeutic cell types requires stringent control of developmental signals, but conventional manual operations introduce substantial intra- and inter-operator variability. While automation can reduce such variability, fixed systems often lack compatibility with the dynamic, stage-specific, and format-diverse manipulations required for hPSC culture and differentiation. Here, we establish a flexible robotic platform to standardize a stroma- and xeno-free embryoid body (EB)-based hematopoietic progenitor cell (HPC) differentiation process and integrate machine learning (ML) to optimize key process parameters including the concentrations of multiple signaling molecules. Despite intrinsic biological stochasticity and technical complexity associated with EB formation, the robotic platform substantially reduced experimental variation. ML-guided, unbiased exploration not only identified combinations of signaling inputs that markedly improved the efficiency and reproducibility of HPC differentiation and subsequent natural killer cell generation, but also illuminated key signaling logic underlying early human hematopoietic development. Together, these results demonstrate that data-driven robotic automation can uncover optimal culture conditions that are difficult to identify through conventional human-driven experimentation.