Artificial Intelligence Technology for Cell Therapy Manufacturing

Regenerative medicine, which will be central to the development of future medical practice, depends upon our ability to correct disease processes by intelligent interventions that either replace diseased tissues or manipulate the disease phenotype using precisely modulated exogenous factors to target specific signalling pathways and regulatory networks.
Pluripotent stem cells are widely recognized as a potential source of specific cell types for use in tissue replacement. They also provide the basis for developing tools for studying disease processes. A major obstacle to achieving the promise of regenerative medicine is our lack of a quantitative understanding of how human pluripotent cells (hPSCs) make choices and control the transition to new fates in response to exogenous stimuli, their position in the colony, the state of the cells in its neighborhood and the endogenous signaling between the cells.
Recent advances in molecular cell biology have demonstrated the feasibility of reprogramming cells I.e. 'switching on/off" the genes that are active in pluripotent cells and those responsible for cell specialisation using transcription factors, small molecules and growth factors. Protocols are available not only for converting embryonic stem cells into specialised cell types but also for reprogramming adult cells into pluripotent state.
Despite the progress made, we have limited knowledge about the trajectories followed by the cell's phenotype during reprogramming procedures and we lack quantitative models that predict how cells' phenotype changes in response to reprogrammed by factors.
Existing reprogramming protocols are cumbersome, slow and produce variable results making it difficult to demonstrate the cell therapy is safe and meets the manufacturing standards for clinical-grade therapies.
Our inability to systematically manipulate cellular phenotype represents a major bottleneck both for the design of new drug and cell replacement therapies as well as in the development of tissue models.
As billions of stem cells are needed just to undertake clinical trials there is a pressing need for high-throughput automated, intelligent cell manufacturing processes that leverage the vast amounts of data generated by laboratories to optimise cell production.
The aim of this project is to apply artificial intelligence technology, predictive modelling and high-dimensional statistical inference to
1- design optimal culture conditions to enrich lineage-specific hESCs subpopulations,
2- discover and quantify robust phenotypic trajectories for differentiation to early germ layers and
3- learn the optimal combination of exogenous factors for directing cells along these trajectories.
A Cellasic microfluidic system, which allows the precise control of the concentration of added factors to the cells, will be used to investigate by real-time microscopy the cells plated in different self-renewal conditions characterized by different proportions of signaling factors (ActivinA, bFGF, Wnt3a BMP) and the dynamics of hESCs when exposed to different perturbations.. Fluorescent intensity of the reporters will be monitored in real time using previously developed in-house segmentation and tracking algorithms. By running experiments in which reporter cell lines (SOX2, GATA6 & OCT4) are simultaneously perturbed, we will generate single cell data to implement deep learning algorithms and develop models that predict the effect of perturbation on cellular fate.