With the tools of computer science and the questions of biology, a team of international researchers was able to uncover the dynamics of complex cellular signaling.
TU Delft computer scientists have applied an optimization technique on large-scale protein measurements in cancerous lung cells in order to help biologists from the United States confirm a theory about chemical signaling in cellular pathways, while also uncovering previously unknown pathways within these cells.
Dr. Marc Grimes, professor of biological sciences at the University of Montana, Missoula, had been investigating his theory that for an individual cell, multi-protein complexes would play a role in integrating several cellular signals into a single instruction for the cell to follow. He found himself with massive spectroscopic measurements of these protein complexes, but because it is common for a mass spectrometer to miss details in complex molecules, his dataset was sparse. Searching for a way to deal with his large and sparse data, he stumbled upon a pattern-recognition-based optimization method that had been developed by Dr. Laurens van der Maaten, an assistant professor in the Pattern Recognition and Bioinformatics Group. “He just sent an email. We’ve never met,” said van der Maaten of the far-flung collaboration.
The optimization work, done by van der Maaten and a former post-doc, Dr. Wan-Jui Lee, enabled the international collaborators to efficiently sift through the high-dimensional proteomic data. The proteins had several variables, or dimensions, and thousands of data points, the proteins. Van der Maaten’s method, called t-Distributed Stochastic Neighbor Embedding, or t-SNE for short, took all of the high-dimensional data as input and reduced the dimensionality by grouping the proteins by how close their variables were to one another. They literally calculated the Euclidean distances among the variables, across all of the data points. The method finally resulted in a two-dimensional visualization that indicated which protein complexes were most similar, facilitating Grimes’s and his colleagues’ interpretation of the data. The team’s findings were published in PLoS ONE in January 2013.
“That our cooperation could happen so fast and smooth [is in] a great part owed to the international working environment of TU Delft,” said Lee, who, from her native Taiwan, first came to TU Delft on a scholarship as a guest researcher and then stayed as a post-doc. “Everyone was open, … brilliant and friendly, she said of her former research group. Lee now applies her research in industry, still in Holland.