AI algorithms empower scientists to design biomolecules with a huge range of valuable functions

April 16, 2022

(News from Nanowerk) When Dr. Shiran Barber-Zucker joined Professor Sarel Fleishman’s lab as a postdoctoral fellow, she chose to pursue an environmental dream: to break down plastic waste into useful chemicals. Nature has clever ways of breaking down hard materials: dead trees, for example, are recycled by white rot fungi, whose enzymes break down wood into nutrients that return to the soil. So why not get the same enzymes to break down man-made waste?

Barber-Zucker’s problem was that these enzymes, called polyvalent peroxidases, are notoriously unstable.

“These natural enzymes are true prima donnas; they are extremely difficult to work with,” says Fleishman, from the Department of Biomolecular Sciences at the Weizmann Institute of Science.

Over the past few years, his lab has developed computational methods that are used by thousands of research teams around the world to engineer enzymes and other proteins with improved stability and additional desired properties.

For such methods to be applied, however, the precise molecular structure of a protein must be known. This usually means that the protein must be stable enough to form crystals, which can be bombarded with X-rays to reveal their structure in 3D. This structure is then modified using algorithms in the lab to design an improved protein that does not exist in nature.

But if the original protein can’t even be produced in the lab or is too fragile to form crystals, as is the case with polyvalent peroxidases, such attempts at improvement can hit a dead end.

Barber-Zucker nevertheless took a chance on the prima donna enzymes, and his timing was odd. Since the 1980s attempts have been made to circumvent the need for crystallization by predicting the 3D structure of a protein from its DNA sequence, but for complex proteins such as peroxidases these predictions were not reliable. Yet, at the end of 2020, several weeks after the launch of his project, the enzyme structures predicted by Barber-Zucker suddenly seemed surprisingly reliable.

It turned out that at this very moment the company from Google DeepMind and several academic research teams had improved artificial intelligence (AI)-based structure prediction methods to the point where they had become highly accurate. This turned out to be a game-changer: the approach led to predicted models that are almost as accurate as those obtained experimentally with crystallography. Armed with the new structures, Barber-Zucker, together with his colleagues – Vladimir Mindel and Jonathan J. Weinstein, research students from Fleishman’s laboratory, and Professor Miguel Alcalde and Dr. Eva García Ruiz from the Institute of Catalysis in Madrid – realized the previous unthinkable (JAC, “Stable and functionally diverse general purpose peroxidases engineered directly from sequences”).

Only one enzyme from the polyvalent peroxidase family had previously been structurally described by researchers, and this project had taken a team of experts about a decade. Now, in less than six months and without any prior expertise in wood-degrading enzymes, Barber-Zucker and his colleagues have succeeded in designing, producing and analyzing stable variants of three versatile peroxidases whose original versions could not, in the past, having been produced in the lab.

The scientists used AI-based 3D models as a starting point. They applied to these models an algorithm created in Fleishman’s lab called Protein Repair One Stop Shop, or PROSS, which engineers a protein modified on the computer to improve its properties on demand.

3D structural model of a versatile peroxidase enzyme generated by an AI-based structure predictor. The yellow dots are the sites of mutations suggested by PROSS to improve the stability of the enzyme

This combined approach opens up a wide range of opportunities.

“Millions of potentially valuable proteins that could not have been accessed biochemically are now within research reach and used in biomedicine and chemistry,” says Fleishman.

It refers to the fact that 3D structures have been solved experimentally for less than 0.05% of the millions of natural proteins whose DNA sequence is known, and that about half of all proteins in nature cannot be effectively expressed and tested in the laboratory.

“These proteins are the dark matter of biology – scientists have no way of pinpointing precisely what they do. In previous studies of protein design, our first question was, ‘Do we have a structure of what protein do we want to focus on? But now that question has become moot, you can get by with or without structure, and that’s a real turning point.

Drug design is an area that could immediately benefit from this breakthrough. For example, antibodies created in laboratory animals must be adapted to humans before they can be used in clinical settings – a laborious process that involves the crystallization and modification of many regions of the animal molecule. The new advance should make this and other antibody engineering processes much more efficient and effective.

Environmental applications, at the origin of this study, are another promising avenue. Wood-degrading enzymes could, for example, be suitable for recycling stubborn agricultural waste. Instead of burning this waste or dissolving it with polluting chemicals, as is often the case today, it may be possible to break it down, using versatile peroxidases, into sugars that can be fermented into biofuel. . Farmers could then recycle in small bioreactors.

Enzymes could also be designed to degrade environmental pollutants. In fact, Barber-Zucker has already shown that its enhanced enzymes can attack a particularly stubborn polluting dye. She also found that each of the three enhanced enzymes exhibited different activity in the lab and that each specialized in breaking down different wood components, suggesting that they may act synergistically.

Importantly, all three enzymes have proven to be remarkably stable and heat resistant, an essential characteristic for their use in industry. Barber-Zucker now aims to develop an enzyme “cocktail” in which a dozen different enzymes, including its versatile peroxidases, will work synergistically to break down wood waste into biofuel or other useful materials.

And what about his vision of recycling hard plastics using these enzymes? “It’s still a dream, but it could come true in the near future,” she says.

Sharon D. Cole