In partnership with the University of Pittsburgh School of Medicine, researchers at MIT have developed a machine learning framework to define the mathematical equations describing a cell’s pathway from one condition to another.
The framework is named “dynamo” and can also determine the underlying mechanisms that drive cell changes. Their research focused on how cells change over time rather than how they migrate through space.
Biological systems are generally difficult to forecast. The researchers stated that they want to follow how a cell changes in response to regulatory gene interactions as precisely as an astronomer can chart the movement of a planet in reaction to gravity and then understand and regulate those changes. They hope to advance single-cell biology to a more quantitative level with their research.
Dynamo creates equations by combining data from many different cells. Achieving this requires how the expression of several genes in a cell changes over time. Because RNA is a measurable result of gene expression, the researchers used changes in the amount of RNA over time to calculate it. They looked at the initial amounts of RNAs and how those RNA levels are changing to anticipate the course of the cell, similar to noting the initial position and velocity of a ball to observe its trajectory.
However, because sequencing only measures RNA once, estimating changes in the amount of RNA from single-cell sequencing data is a difficult task. Therefore, it became essential for the team to estimate how RNA levels changed using clues such as RNA being generated at the time of sequencing and RNA turnover equations.
The team built upon previous approaches to generate clean enough data for the Dynamo to operate. They used a recently developed experimental method that tags new RNA to distinguish it from old RNA with sophisticated mathematical modeling.
Next, they tried viewing cells at discrete moments in time to a continuous picture of how cells change. Methods for comprehensively profiling transcriptomes and other ‘-omic’ information with single-cell resolution have advanced tremendously. Yet, the analytical methods for exploring these data have been descriptive rather than predictive. Therefore, they employed machine learning to uncover continuous functions that describe these spaces. By converting these functions into math-based maps, Dynamo can visualize them. The present gene expression dynamics determine a cell’s beginning position on the map. Later, trails from where the cell begins can be followed to know where it will end.
The researchers tested Dynamo’s cell fate predictions on cloned cells. The findings show that the sequence of one of two nearly identical clones would be performed while the other clone differentiated. Dynamo predicted that each sequenced cell would happen to match what transpired to its clone.
The team used blood cells to evaluate their approach. They discovered that Dynamo correctly recorded blood cell development and verified a recent observation that megakaryocytes form earlier than other types of blood cells. Furthermore, Dynamo was also able to reveal the mechanism underpinning this early differentiation.
The researchers state that their proposed framework will not only aid with their understanding of how cells migrate from one state to another but will also assist in managing this shift. To that purpose, Dynamo offers tools for simulating how cells will change in response to various perturbations. In addition, it provides a mechanism for determining the most efficient path from one cell state to the next. These techniques give a robust foundation for predicting how to best reprogram any cell type to another, a major challenge in stem cell biology and regenerative medicine, as well as generating ideas about how other genetic modifications may affect cells’ fate.