How Biologists are Using Machine Learning to Study Cells

In a field such as astrophysics, scientists use machine learning (artificial intelligence) to make sense of the infinite — such as stitching together signals from telescopes scattered around earth to map and image a distant black hole.

In biology, researchers are applying the same software and hardware advances to answer questions at the cellular level. Here are three inquiries made possible by machine learning.

Can we model cell architecture and discover new cell parts?

Researchers at University of California San Diego School of Medicine, Stanford University and KTH Royal Institute of Technology in Stockholm, Sweden asked a fairly basic question: could we create a better map of human cell architecture?

Studying a human kidney cell line, the researchers found 69 subcellular systems — half of which had never been seen before. In one example, the researchers spotted a group of proteins forming a new complex of proteins that binds RNA. The complex is likely involved in splicing, an important cellular event that enables the translation of genes to proteins, and helps determine which genes are activated at which times.

Existing microscopes and biochemistry techniques produce important data. The computing power of machine learning can be trained to turn that data into a more complete model of the cell. Scientists hope to repeat the process with all human cells — healthy and diseased — and then compare. Such detail should reveal patterns key to understanding the molecular basis of many diseases.

Can we model cell behavior with more accuracy?

The key to understanding human cell behavior is to label as many of the parts as possible, and then watch how the parts interact over time.

Traditional fluorescence microscopy offers scientists a limited range of colored labels, and requires expensive reagents that damage the very cell scientists are trying to study. One alternative is called transmitted white light (bright-field microscopy), a technique that avoids the stains, but settles for a lower contrast, meaning most cell structures remain too difficult to spot.

Scientists from the Allen Institute for Cell Science in Seattle, Washington applied the computing power of deep learning to get the best of the two techniques. Artificial intelligence (AI) refined the bright-field images by applying a ‘virtual staining’ — allowing a better look inside the cell without the damaging effects of Traditional fluorescence microscopy.”

Can we predict how our cells will respond to therapeutic changes?

The great hope of regenerative medicine is the idea that our understanding of human cells might one day be so complete that drugs and other therapeutics can be personalized at the cellular level.

For now, science recognizes that differences happen from cell to cell and person to person. Scientists have a hard time predicting how cells will respond in any given therapeutic scenario.

Simon and his colleagues want to change that by narrowing down the possibilities and increasing the chances the doctor will know that these cells will do exactly what’s expected.

Scientists from the National Institute of Standards and Technology and the National Eye Institute at the National Institutes of Health turned to deep learning to better match lab-grown eye tissue to a range of patients suffering from macular degeneration. 

While humans struggle with the detailed observations and mathematical calculations required to compare terabytes of microscopic data, the machine learning tools proved to be relatively fast and accurate. John Carroll University’s Cell and Molecular Biology degree helps its students develop the skills needed to engage with exciting research opportunities such as this.