Could the same approach that mapped the Internet be used to identify tumor cells? Bulent Yener, who has devoted more than a decade of research to the idea, recently reviewed how his work and that of other researchers contributed to biomedical research in “Cell-Graphs: Image-Driven Modeling of Structure-Function Relationships,” published in the January edition of Communications of the Association of Computing Machinery. The above video, which accompanies the article, explains how Yener and other researchers used an unorthodox analysis of the interactions between cells to determine their function.
In the video, Yener, a Rensselaer professor of computer science and director of the Data Science Research Center, explains how he transferred the techniques he applied in working on a map of the Internet produced by Bell Labs in 1999 to systems biology:
I asked myself, what would grow like the Internet, in a selfish, decentralized, chaotic way? Maybe cancer has similar behavior. Can I build the map of a tumor, the graph of a tumor, can I show that graph is different than a tissue without a tumor. And that’s how it started.
Yener’s work, funded by the National Institutes of Health, begins with images of cells taken from a sample of tissue or an organ. By applying a variety of image processing techniques, researchers are able to obtain information about multiple aspects of the structural organization of the cells. Then, the cell graph technique uses graphing theory to determine the structure-function relationship of the cells by modeling the structural organization of that tissue/organ sample. Here’s how Yener describes the approach in the CACM article:
Its main hypothesis is that cells in a tissue/organ organize to perform a specific function. For example, the spatial distribution and interaction of cells in a salivary gland tissue is different than that of brain tissue since they perform very different functions. Thus, if one can understand tissue organization then one can successfully predict the corresponding function. The cell-graph technique deploys image processing, feature extraction and selection, and machine learning algorithms to establish a quantitative relationship between structure and function.
In a cell-graph, nodes represent the cell nuclei and pairs of nodes are connected by a link based on the spatial, chemical, or biological relationship between them. From this information, Yener creates a matrix, or graph, that represents the links between the nodes. In his vernacular, nodes are also called “vertices,” and links are also called “edges.”
Once I have this matrix, in this form, then I can do magic using math. From this matrix I can calculate more than 120 properties, or as we call them ‘features’ that are encoded in the table about the spatial relationships of these nodes or vertices. That’s where the power of graph theory comes from.
The properties Yener calculates denote the structural organization of the cells in the tissue or organ, and by applying various machine learning techniques to that information, he can classify, predict, or diagnose the functional state of the tissue based on that organization.
It’s a tool that makes it possible for people to discover existing relationships that are hidden to us.
Graph theory could be used to map any kind of complex relationship with multiple players. In his own work, Yener is enabling a shift from a reductionist approach – that dissects the system and then tries to put each piece together – to one based in computation.