Controlled iPSC positioning through machine learning

Written by Alexander Marshall

Researchers have used a predictive software to control the orientation and layout of iPSCs.

An algorithm designed to predict iPSC organization has allowed researchers from the Gladstone Institutes (CA, USA) to control the arrangement and growth of their cells in vitro. Spatial organization is thought to be a major variable in determining how iPSCs differentiate, so researchers hope that being able to manipulate their positioning will enable better control of differentiation and move towards ex vivo organ engineering.

By manipulating two genetic pathways, the researchers — in collaboration with their colleagues at Boston University (MA, USA) — were able to create a limited set of spatial alterations that could be utilized by machine-learning to establish a predictive program. The team was then able to utilize the program in reverse, computing the parameters necessary for various shapes. When these conditions were applied to their iPSCs, the team were able to produce elaborate patterns such as small islands and bullseyes, formed from concentric circles.

“I was just blown away when I first saw the results,” explained Bruce Conklin (Gladstone Institutes). “Modeling cell behavior is the Holy Grail of biology and this paper takes an important step forward in doing that.”

Normal stem cells have an innate ability to position and localize themselves, allowing them to develop into complex organs during development; however, iPSCs have shown disorganized structures, which some have tried to overcome with three-dimensional stem cell printing. Unfortunately, this is not always successful, resulting in researchers searching for a means to control the iPSC positioning biologically.

Previously, the team has demonstrated the substantial influence of ROCK1 and CDH1 on cell positioning, altering the cells’ orientations by several degrees depending on the length and strength of inhibition and the proportion of cells affected. By knocking down the expression using CRISPR-Cas9, the researchers were able to collect data which they could then pass to the machine-learning algorithm, allowing patterns to be discerned.

The team hope that they will be able to add other variables to the predictive program, incorporating different developmental genes, with the aim of eventually advancing towards the capability of directing cells to form three-dimensional constructs.

Sources: Libby A, Briers D, Haghighi I, et al. Automated Design of Pluripotent Stem Cell Self-Organization. Cell Syst. doi: 10.1016/j.cels.2019.10.008 (2019) (Epub ahead of print); https://gladstone.org/about-us/news/machine-meet-stem-cells

Have any additional questions about this story? Ask us in the comments, below.


Find out more in these top picks from the Editor: