A new artificial intelligence method shows that we don’t know half of what’s in our cells

Most human diseases can be traced back to the malfunction of a part of the cell-for example, a tumor may grow because a gene has not been precisely translated into a certain protein or a metabolic disease occurs because the mitochondria are not working properly. But to understand which parts of a cell might fail in a disease, scientists first need to have a complete list of parts.

By combining microscopy, biochemistry techniques and artificial intelligence, researchers from the University of California San Diego School of Medicine and their co-authors have made what they believe is a significant leap forward in understanding human cells.

The technique, known as Multi-Scale Integrated Cell (MuSIC), is described Nov. 24, 2021, in the journal Nature.

“If you imagine a cell, you probably draw a colorful diagram from a cell biology textbook with mitochondria, endoplasmic reticulum and nucleus. But is that the whole story? Definitely not,” says Trey Ideker, Ph.D., professor at the University of California San Diego School of Medicine and the Moores Cancer Center.

“Scientists have long realized that we don’t know more than we know, but now we finally have an opportunity to look deeper.”

Ideker led the study with Emma Lundberg, PhD, of the Royal Institute of Technology KTH in Stockholm, Sweden, and Stanford University.

In the experimental study, MuSIC identified about 70 components found in a human kidney cell line, half of which had not been seen before. In one example, the researchers noticed a group of proteins forming an unfamiliar structure.

In collaboration with colleague Gene Yeo, Ph.D., of the University of San Diego, they determined that this structure was a novel complex of proteins that bind RNA. This complex is probably involved in splicing, an important cellular event that ensures the translation of genes into proteins and helps determine which genes are activated at a particular time.

The internal structure of cells and the many proteins within them are usually studied using one of two methods: microscope imaging or biophysical association. In imaging, researchers add fluorescent tags of different colors to the proteins of interest and track their movement and association in the microscope field of view. To study biophysical associations, researchers can use an antibody specific to a protein to remove it from the cell and see what else is attached to it.

The team has been interested in mapping the inner workings of cells for years. What makes MuSIC different is its use of deep learning to map the cell directly from cell microscopy images.

Left: Classic textbook cell diagrams imply all parts are clearly visible and defined. Right: A new cell map generated by MuSIC technic reveals many novel components. Gold nodes represent known cell components, purple nodes represent new components. The size of node reflects number of distinct proteins in that component. Credit: UC San Diego Health Sciences

“The combination of these technologies is unique and effective because it’s the first time we’ve been able to combine measurements at completely different scales,” said study first author Yue Qin, a graduate student in the Department of Bioinformatics and Systems Biology in the Ideker lab.

Microscopes allow scientists to see down to the one-micron level, which is roughly the size of some organelles, such as mitochondria. Smaller elements, such as individual proteins and protein complexes, are not visible under the microscope. Biochemistry techniques that start with a single protein allow scientists to get to the nanometer scale.

“But how do you bridge that gap from the nanometer to the micron scale? That has long been a big hurdle in the biological sciences,” says Ideker, who is also the founder of the Cancer Cell Mapping Initiative at the University of California and the University of California, San Diego’s Center for Computational Biology and Bioinformatics.

“It turns out that this can be done with artificial intelligence — examining data from multiple sources and asking the system to assemble it into a cell model.”

The team trained the MuSIC artificial intelligence platform to look at all the data and build a cell model. The system does not yet tie the contents of the cell to specific locations, as in the diagram in the textbook, in part because their locations are not necessarily fixed. On the contrary, the arrangement of the components is fluid and changes depending on the type of cell and the situation.

Ideker noted that this was a pilot study to test MuSIC. They only looked at 661 proteins and one cell type.

“Obviously, the next step is to study the whole human cell,” Ideker said, “and then move on to different cell types, people and species. Eventually we will be able to better understand the molecular basis of many diseases by comparing the differences between healthy and diseased cells.”

Notify of
Inline Feedbacks
View all comments
Would love your thoughts, please comment.x