The worlds of artificial intelligence (AI) and biology are becoming increasingly intertwined. We’ve seen some pretty remarkable things grow out of synthetic biology: lab-grown meat, recyclable biofuel, and super-powered yeast cells that create life-saving drugs like insulin. A subfield of biology, synthetic biology, often gets down and dirty with genomics and gene editing to develop something entirely new.
But genomics and gene editing take a lot of time, effort, and experimentation. So far, synthetic biologists have largely taken a trial-and-error approach to understand the impact of a gene change in a cell. By marrying synthetic biology to AI and machine learning applications, however, we could greatly reduce the time and cost to develop a new product, while mostly eliminating the trial-and-error approach.
This past month, a team of researchers at the Department of Energy’s Lawrence Berkeley National Laboratory, located a 30-minute drive from San Francisco, successfully tried a different approach. The team, led by Dr. Hector Garcia Martin, used machine learning to predict how a cell’s biochemistry and behavior are affected by changes to the cell’s genes. Instead of trial-and-erroring through a multitude of manipulation combinations, the researchers developed a machine learning algorithm that offered recommendations on the next bioengineering cycle.
The algorithm also provided predictions on how different changes to the cell’s genes would lead to a specific goal. This allowed the team to look ahead before wasting time and effort by actually going through the cycles. According to Martin, “If you’re able to create new cells to specification in a couple weeks or months instead of years, you could really revolutionize what you can do with bioengineering.” He added that the “possibilities are revolutionary.”
Synthetic biology was named one of the top ten emerging technologies by the World Economic Forum in 2016 because it has the potential to change the world as we know it. It can make mosquitos obsolete with gene changes or boost agricultural growth with robust microbiomes while reducing the consumption of environmentally-unsafe fertilizers. In synthetic biology, there are several subfields that branch out, but the subfield with the most eyes on it right now is metabolic engineering.
Everything that’s living requires metabolism; it uses cells to transform food into energy and energy storage (fat). Synthetic biology is a very similar analogy: it changes a cell so that the cell takes in something and transforms it into something else. For example, a yeast cell gets changed from a simple cell with no idea of blood sugar to pumping out insulin.
However, because cells are so complex, it’s not known exactly what the implications of manipulating the cell are without running experiments. But with historical data, AI applications can “predict biological system behavior,” according to the research team. Besides the predictive nature of this algorithm, the application can extract trends from the experimental data and look much further ahead than even the most experienced lab researchers.
The Power of Probability
The team’s algorithm is called ART (Automated Recommendation Tool). It uses probability and statistics to make predictions based on what it’s already learned. At the foundation of the algorithm is Bayesian statistics, which is commonly used in machine learning. Although machine learning algorithms require thousands or more rows of data, it’s not realistic for the field of synthetic biology where large datasets are rare.
To compensate, the team has tweaked their algorithm to work with a small amount of data, and the algorithm thrives off of the uncertainty with a good level of precision. Currently, the algorithm learns from proteomics, which is a list of all proteins in a cell, to develop a probabilistic model that predicts how changes to the cell’s genes will affect the rest of the experiment. ART can work with a specific cell as the goal or it can simply work off of the desired outcome like “decrease production of an unwanted biochemical” or “increase the amount of desired biochemicals”.
Throughout each step of the prediction process, the algorithm also tells the team the levels of the biochemical so that the experiment can be adjusted in advance to develop the desired chemical.
The Right Path
To get the algorithm to this level of success, the team tested ART five times with different types of datasets. For example, one test involved asking ART to use living cells to optimize carbon-neutral biofuel production. The algorithm was trained on 27 different biological pathways beforehand, and the result was a synthetic pathway that’s automatic and efficient.
Although ART’s predictions for a given fuel chemical goal weren’t the most accurate, it always pointed the researchers in the right direction to improve production. The team says that, somehow, ART learned “the gist” of what biofuel manufacturing is. Like humans, ART doesn’t fully understand biochemical pathways and their underlying mechanisms, but it gets it enough to save researchers copious amounts of time.
Recently, the team used ART to brew hoppy beer without using hops. The algorithm pointed the researchers to bioengineered yeast that was programmed to brew ethanol. The yeast was then manipulated to synthesize chemicals that give a hoppy flavor. The team says that growing hops requires a lot of energy and water, so the taste isn’t consistent in beer, even if the hops came from the same farm. Hops-free beer could help the environment while keeping us drinking familiar-tasting beers.
In another experiment, ART programmatically looked through 8,000 combinations of biochemical pathways to create tryptophan (known for making you sleepy after Thanksgiving dinner), complete with recommendations on how to double the chemical’s production.
The Technology of Tomorrow
Machine learning may be the secret ingredient that accelerates the innovation, growth, and progress of the synthetic biology industry. Martin has hope that ART can be used to bioengineer “the full genome”, and one day, even “automate metabolic engineering.” With AI and machine learning, anything is possible.
Tell us — would you try a hops-free beer at a local brewery? Let us know in the comments below!Tags: AI and beer, AI app developer San Francisco, app developers san francisco, artificial intelligence app development, machine learning and beer, machine learning app development San Francisco, machine learning apps, mobile app developer San Francisco, San Francisco AI app developers, San Francisco app development, San Francisco machine learning app developer, San Francisco mobile app development