Scientists at the University of Tokyo Institute of Industrial Science have designed machine learning algorithms to predict the size of individual cells as they grow and divide. By using artificial neural networks that do not impose assumptions commonly used in biology, computers were able to make more complex and accurate predictions than ever before. This work may help not only advance the field of quantitative biology, but also improve the industrial production of pharmaceuticals and fermented products.
As with all natural sciences, biology has developed mathematical models that help fit data and make predictions about the future. However, due to the inherent complexity of biological systems, many of these equations rely on simplified assumptions that do not always reflect the actual underlying biological processes. Researchers at the Institute of Industrial Science, the University of Tokyo are currently implementing machine learning algorithms that can predict future sizes using single cell sizes measured over time. The computer automatically recognizes patterns of data, so it is not as constrained as traditional methods.
“In biology, simple models are often used based on their ability to reproduce measured data,” says lead author Atsushi Kamimura. “But the model may not be able to capture what is really happening because of human prejudice.”
Data from this latest study were collected from either Escherichia coli bacteria or Schizosaccharomyces pombe yeast cells retained in microfluidic channels at various temperatures. The plot of size over time looked “sawtooth” because exponential growth was interrupted by a division event. Human biologists typically use the “sizar” model, which is based on the absolute size of cells, or the “adder” model, which is based on the increase in size since birth, to predict when division will occur. Computer algorithms have found support for the “adder” principle, but as part of the complex web of biochemical reactions and signal transduction.
“Our deep learning neural networks can effectively separate history-dependent deterministic factors from the noise of a given data,” says senior author Tetsuya Kobayashi.
In addition to predicting cell size, this method can be extended to many other aspects of biology. In the future, life sciences may be driven by more objective artificial intelligence than human models. This has the potential to more efficiently control the microorganisms used to ferment products and manufacture pharmaceuticals.
material Provided by Institute of Industrial Science, University of Tokyo.. Note: The content can be edited in style and length.
Decoding Noisy Data to Predict Cell Proliferation-ScienceDaily
Source link Decoding Noisy Data to Predict Cell Proliferation-ScienceDaily