Advancements in sequencing, cytometry, and imaging technologies can now provide an unprecedented level of detail for characterizing immune cells. The advent of these novel technologies, such as mass cytometry and single-cell RNA sequencing, brings a systems-level view for investigating the immune system. However, these technologies produce large amounts of high-dimensional data, making it challenging to analyze. Traditional manual data analysis of high-dimensional data is time consuming, subjective and difficult to reproduce, as human error and bias accumulate in increasing dimensionality. As further advancements are made, artificial intelligence methods coupled with exciting new technologies will soon transform medical research.

Single-cell technologies can now produce highly complex data sets with unparalleled resolution. One such technology, mass cytometry by time of flight (CyTOF), can measure over 50 surface and intracellular markers per cell. CyTOF uses rare metal isotopes as reporters conjugated to antibodies, measuring marker expression by detection of metal abundance. It will be possible for CyTOF to measure over 200 markers on single-cells as new metal -conjugated antibodies are developed. More recently, CyTOF has been adapted for imaging tissues with subcellular resolution, providing spatial information and drastically increasing dimensionality of the data. State of the art artificial intelligence techniques for image recognition will be fundamental in analyzing imaging mass cytometry (IMC) data. Similarly, single-cell RNA sequencing (scRNA-seq) technologies produce complex high-dimensional transcriptomic profiles, measuring 5,000 to 10,000 genes per single-cell. These technologies provide unprecedented and overwhelming resolution, making data analysis extremely challenging. With traditional data analysis methods, it is nearly impossible to fully utilize the depth of information captured by these technologies. Oftentimes, data is reduced to lower dimensionality before analysis, using algorithms such as principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE).

Recent establishment of the Vector Institute, Toronto’s institute for artificial intelligence research, creates exciting opportunities for collaboration and new applications of the most cutting-edge deep learning methods in medical research. Deep learning is a class of machine learning that uses artificial neural networks inspired by structural and anatomical research of the brain. Deep artificial neural networks have made leaps and bounds in language processing, speech recognition and image recognition within just the last decade. University of Toronto professor Dr. Geoffrey Hinton, the “godfather of artificial intelligence”, now leads the Vector Institute as Chief Scientific Advisor. The most sophisticated image recognition, speech recognition and general intelligence systems, such as Google Translate and AlphaGo, stem from Hinton’s work on deep artificial neural networks.

The Institute will receive support from Ottawa and Ontario, a combined investment of $100 million, with additional investment of over $80 millionrobot-507811_1920 from more than 30 private sector companies. The Vector Institute will help Toronto to become a world-leading centre for artificial intelligence research, fostering new opportunities for collaboration and economic growth. Machine learning is crucial for single-cell data analysis. In fact, one of the most important machine learning algorithms for analyzing CyTOF and scRNAseq datasets, called t-SNE, was developed by Hinton and Dr. Laurens van der Maaten. t-SNE is a non-linear dimensionality reduction algorithm, mapping high dimensional data onto a two-dimensional plot for visualization. The main application of t-SNE, before it was adapted for CyTOF and scRNA-seq technologies, was to visualize high-dimensional abstractions learned by deep neural networks.

Integration of deep learning in advanced medical research technologies will be revolutionary. Measuring over 50 markers, CyTOF could be used to identify hematological disease subtypes and determine prognostic/diagnostic outcomes, providing patients with more personalized treatments. Training deep neural networks for determining prognostic/diagnostic outcomes with CyTOF is one possible future application for deep learning. Further exploration and development of new deep neural networks specific for biological applications will be critical for fully utilizing technologies such as CyTOF, IMC and scRNA-seq. It is an incredible and exciting time for medical research with the rise of artificial intelligence and advancements in single-cell technologies.

1. Allen K. New institute aims to make Toronto an ‘intellectual centre’ of AI capability. Toronto Star (2017).
2. Auld D. Imaging Mass Cytometry. Genetic Engineering & Biotechnology news (2014).
3. Berkowitz R. Identifying immune cell subsets with CyTOF. The Scientist (2017).
4. Smith C.S. The Man Who Helped Turn Toronto Into a High-Tech Hotbed. The New York Times (2017).
5. Svensson V. et al. Power analysis of single-cell RNA-sequencing experiments. Nature Methods. 14:381 – 387 (2017).
The following two tabs change content below.

Anthony Zhao

Previous post Science & Politics: An Intertwined Web
Next post Is Bigger Data Better Data?

One thought on “Artificial Intelligence: The Future of Medical Research

Leave a Reply

Your email address will not be published. Required fields are marked *


Feed currently unavailable. Check us out on Twitter @immpressmag for more.