Report | The tool revolutionising biotech by predicting protein structures

This summer, an article in Nature shook the world. A new computational tool with a public database was launched, offering for the first time the structure of 365,000 proteins from 21 model organisms, from bacteria and plants to brewer’s yeast, mice and Homo sapiens; all chosen based on biotechnology criteria as they are the most commonly used in research. The science community understood that this would be a turning point in studying proteins. Alpha Fold2, created by British AI company DeepMind, which was acquired by Google, seems to resolve a problem that has been troubling biology for the past 50 years. This advance opens up a new era in obtaining protein structures that could, in the near future, bring huge improvements to the health and wellbeing of people and the planet given its impact on drug discovery, plant defences and breaking down plastics, among the many other applications imaginable. 

Companies like Zymvol and ProtoQSAR, and research centres like the Centre for Plant Biotechnology and Genomics (CBGP [UPM-INIA]) already have an eye on its development. They shared with us how it has been received and how it could affect biotechnology.

For many biotechnology processes, understanding what is going on at a molecular level is key, and so protein structures are essential both for analysing the impact of a pathogen on plants and discovering new drugs, and for treatments. Faced with this need, all the specialists we consulted agreed that AlphaFold2 is a powerful advance in this field, as it can predict protein structures computationally with accuracy equivalent to experimental work and has already provided protein structure models that had never been predicted before. “The structures of the immense variety of proteins that are of interest to the biotechnology sector weren’t available and this was a huge limitation for all of us, hindering the accuracy of our prediction work. AlphaFold2 has changed all that”, noted Luis Fernández Pacios, professor at the Polytechnic University of Madrid and CBGP researcher. 

This tool will facilitate, speed up and open new doors in biotechnology research. “Its biggest advantage, above all, is the quality of the three-dimensional protein structure. It will reduce the risk of errors”, explained Ferran Sancho, senior researcher at Zymvol, a company that specialises in computer-aided enzyme discovery and engineering. Thanks to this new advance in artificial intelligence, Zymvol will be able to create more and better models, which will allow them to work with clients in new areas and applications that it hadn’t considered previously due to technical limitations. “I think this is the most impactful news I’ve heard since I went into science”, he said. “It is going to have a huge impact in a few years and unlock many projects and lines of research, but we’re still in the testing phase”, he added. 

Limitations and challenges

Although everyone agrees it is a ‘revolutionary’ tool for the 21st century, there are some limitations we have to look at. First of all, its collection of proteomes needs to be expanded to include associated structures. This is just the beginning, “an introduction”, Pacios said, but now it will be a question of adding more and more information and filling in the gaps by learning. 

At ProtoQSAR, they recognize the tool’s potential to fill in those gaps where there isn’t any information, but they also see some important barriers. Proteins are dynamic structures that don’t have a specific shape throughout their life and that interact, sometimes without order, with many other complexes that can affect them. “For example, if you just start with a sequence, you lose out on all the information on the environment the protein exists in. AlphaFold2 only predicts the structure, which falls a bit short...”, warned Stephen Jones Barigye, the company’s Chief Scientific Officer in a video call from Canada. Another relevant point for this expert is that the tool doesn’t allow you to study mutations, either. “Nevertheless, it still has numerous pluses. It allows us to combine techniques to yield much more reliable results. Now we have something stronger to start with and that’s very valid”, he recognized. 

Another of the barriers mentioned in the interviews was that the AlphaFold2 method doesn’t work with all the proteins imaginable. Fernández Pacios explained that, if the protein sequence being used for the biotechnology process isn’t related closely enough to other known proteins, the tool doesn’t work nearly as well. According to its creators, the accuracy of the predicted structure drops notably if the protein sequence studied doesn’t have at least 30 sequences to be identified from. “There is a lot of room for improvement in filling in those gaps for less common proteins because almost any organism can have hundreds”, he assumed. 

And now what?

What does this tool’s improvement hinge on? Will we one day be able to use a full platform and fill in all those gaps? The experts assured us that this is one of the challenges and that we’ll see many changes and improvements over the coming years. Rafael Gozalbes, CEO of ProtoQSAR, remembered when computational models were first being developed, it seemed like “the computers were going to discover new drugs and we wouldn’t need labs any more”. But no, this specialist insisted, computing is an additional tool, and we’ll always need others to complete the puzzle. “It lets us work better and faster, that’s true”. 

The same company’s Business Development manager, Simón Perera, insisted on the importance of promoting its nationally and in Europe in order to complete and improve the tool without depending on third parties. “We also need to get up-to-date internally, so we don’t miss out on this great improvement in quality”, he noted. 

In any case, all the experts we spoke with trust that this new tool that has been making the rounds of the biotech industry in recent months is going to unlock a whole world that was previously closed, in the breakdown of molecules and plastics, protecting flora, treatments for cancer and discovering new drugs. 

                    —   Are we ready for this revolution in Spain? 

                    —  “We have very good staff, but that’s not enough. The industry has to be aware that this tool and AI are now a permanent part of all our scientific activities, and that we are going to need experts in this field”, concluded Pacios. 

By Agathe Cortes and Raquel Álvarez