Researchers have developed an algorithm known as ART - Automated Recommendation Tool. Art received an enormous amount of information about the set of proteins in the cell, and as a result he could predict how changes in those proteins would affect the substances that the cell is supposed to create
Leave the computer for a moment and go to the balcony of your house. Don't forget to come back to continue reading the entry, but while you're there, look around you and notice the trees surrounding your house. These may be pine or cypress trees, fir or (if you're lucky) apples or oranges. Either way, one thing characterizes them: they all grew from one tiny seed, were helped by the resources around them - soil, water, air and sun - and were shaped so that they could get the most out of all of these. And despite the difference between one pine tree and another, it is clear to all of us that it is still a tree of the same species. That's how accurate the algorithm that directs their growth is.
This is one of the most common wonders on earth. So common that we don't notice it - no more than we marvel at the fact that a perfect human embryo develops from a single cell. Even so, it is a wonder, and it can also teach us about the power of biology in creating machines from scratch. From just one cell, programmed in the right way, we can get the wonderful machine that is the human body, or an infinite number of other wonderful machines: a shark, a whale, a tree, and everything in between.
But these are visions that are still very far from us. Meanwhile, we encounter difficulties even in programming individual cells to produce certain substances necessary for man. It took many years, for example, until researchers were able to engineer microorganisms to produce human insulin. Even today, it takes years to engineer algae in order to improve their biofuel production capacity, or to genetically engineer cells to deal with genetic diseases.
Attempts were made in the past to reprogram cells through genetic engineering, but it turned out to the researchers that this was not a simple or easy task. In each human cell there are more than twenty-thousand genes. Each of them can be 'on', 'off' or any level in between. The products of many of these genes affect the original genes themselves, or other genes, or other substances in the cell. If we want to understand the effect of a game on a certain gene, we will have to calculate its effect on another twenty-thousand other genes and countless other proteins... each of which can affect those twenty-thousand genes, each of which can affect twenty-thousand genes, which One of them can affect…
You got it.
A computational challenge that until now only supercomputers could handle
This is a computational challenge at scales that even today's most advanced supercomputers are unable to solve. And this is only the first challenge on the way to creating pre-programmed organisms. And even that we are unable to solve, without many years of trial and error just to change a gene or two in the cells and bring about the desired effect.
At least, that has been the case so far. But research published in recent months provides new hope for the field: hope that artificial intelligence advanced enough can solve this dilemma and explain to scientists exactly how to reprogram the cells to achieve the desired result.
The research, a product of the laboratory of Dr. Hector Garcia Martin, was published in the respected scientific journal Nature. And its meanings can be summed up in the following words of Martin -
"The possibilities are endless. Currently, bioengineering is a very slow process. It took 150 man-years to create the drug artemisinin against malaria. If you can make new cells on demand in a matter of weeks or months instead of years, you could really revolutionize what you can do with bioengineering.”[1]
The artificial intelligence that does not understand biology
Something strange is happening in the realms of artificial intelligence. In recent years we have witnessed the emergence of artificial intelligence engines capable of receiving large amounts of information, processing them and providing predictions for the future in fields related to that information. These engines do not 'understand' the meaning of the information, or create a detailed theory or model as humans do. They simply learn from millions of cases, on a statistical level, what is most likely to happen.
The researchers in the study in question developed a similar algorithm known as ART - Automated Recommendation Tool. Or in Hebrew - simply Art. Art received an enormous amount of information about the set of proteins in the cell, and as a result he could predict how changes in those proteins would affect the substances the cell was supposed to create. He could, for example, predict what changes are required in a cell in order to increase the production of certain substances, such as drugs and biofuels. Alternatively, it could also explain to researchers how to suppress the production of unwanted substances such as toxins. He could even predict the levels of substances the cell produced - an extremely difficult task.
Some of Art's successes demonstrated in the study include, for example, the optimization of a biofuel production process by living cells. The algorithm relied on information that came from 27 previous biological pathways, and thereby decoded a new artificial metabolic pathway, which was more efficient than what currently exists - and by the way, it was also automatic and did not require special operation of the cells to continue producing the biofuel.
Not everything is rosy: Art's predictions were not entirely accurate, but they always pointed researchers in the right direction to improve production. It is not clear how, but Art 'understood' the way in which cells produce different substances, and provided the researchers with the recommendations for new and promising research directions.
As another example of Art's abilities, the researchers harnessed him to improve the way yeast produces alcohol. The researchers re-engineered the yeast according to Art's recommendations, in order to produce substances that imitate the taste of hops - a plant used as an important ingredient in the beer industry. Growing hops requires huge amounts of water and energy, so its production by the yeast should lower the price of beer - at least the one that contains hops. Art even envisioned the best level of flavoring production to produce a perfect beer (Pale Ale, if you're interested).
"This is clear proof that machine learning leading to bioengineering is possible, and disruptive if it can be applied on a large scale. We did this for five genes, but we believe it is possible to do this for the whole genome.” Garcia Martin said. "This is just the beginning."
[1] https://www.eurekalert.org/pub_releases/2020-09/dbnl-mlt092320.php
