A cure for cancer? Fighting the climate crisis? Ask the data scientists

The days of the “IT crowd” banished to the basement are long gone. Take bioinformaticians, for example – they’ve moved up to the top floors and become an integral part of research teams. They’re well versed in AI tools and know their way around data. “Thanks to them, we’re able to better predict and tackle threats like climate change or global epidemics such as COVID-19,” says Jiří Vondrášek from the Institute of Organic Chemistry and Biochemistry of the CAS, who heads the Czech branch of the ELIXIR infrastructure for bioinformatics.

The Nobel Prizes in Chemistry and Physics 2024 go to… artificial intelligence. That’s how last year’s decision of the Royal Swedish Academy of Sciences could be summed up in a nutshell. In both scientific fields, the prestigious award went to researchers who pushed the boundaries of AI and its applications in science. The Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton for developing machine learning based on artificial neural networks. Thanks to their work, we now take for granted things like translation apps on our phones – or getting answers to all-encompassing questions from ChatGPT.

But the story of the Nobel Prize in Chemistry 2024 is perhaps even more intriguing. It went to biochemist and bioinformatician David Baker for his contributions to designing new proteins, and to computer scientists Demis Hassabis and John Jumper from the tech company DeepMind for the computational tool and AI model AlphaFold. The Nobel Committee’s decision made it crystal clear: experts in IT (information technology) not only belong in science, but have enormous potential to drive it forward by leaps and bounds.

AlphaFold is an AI model that can predict, with astonishing accuracy, how a protein will fold – based solely on the sequence of its amino acids.

A data revolution in biology
Talk to any bioinformatician about their field, and sooner or later AlphaFold will come up. That’s because it really is a revolutionary tool, one that in just a few short years has managed to speed up and refine the determination of existing protein structures and made it possible to design entirely new proteins.

A protein is a molecule that consists of a chain of amino acids. It’s one of the basic building blocks of all living things. Each protein has its own specific function, which depends on how the amino acid chain folds in three-dimensional space. For decades, scientists struggled to figure out how nature produces these structures. In the technical jargon, the challenge became known as the “protein folding problem.”

Gradually, biologists uncovered one protein structure after another – at first mainly thanks to methods like X-ray diffraction (determining atomic composition using X-rays) and nuclear magnetic resonance (NMR). Crucially, the results of these observations and experiments were recorded in databases (such as PDB or UniProt) that are freely accessible.

Proteins are the essential building blocks of all living organisms. They consist of amino acids and perform key functions such as building muscle, supporting immunity, and regulating hormones.

Decades of work have yielded information on more than 200 million protein sequences and 200,000 protein structures. That’s an enormous amount of data – far too much for the human brain to handle, but perfect fuel for artificial intelligence tools.

Feeding AI – the right way
“The breakthrough in predicting protein structures was possible only thanks to these databases. Without the public data that scientists painstakingly collected and then voluntarily shared with the global research community, tools like AlphaFold would never have come into existence,” explains Jiří Vondrášek from the Institute of Organic Chemistry and Biochemistry of the CAS.

Robust datasets serve as training material for AI, which uses them to fine-tune its algorithms. But in order to get good output from AI tools, you have to feed them quality input. “These days, you often hear about the concept of ‘garbage in, garbage out.’ What that means is that no model or system will ever be smarter than its inputs. In other words – even the best AI will spew nonsense if you feed it bad data,” Vondrášek adds.

Jiří Vondrášek from the Institute of Organic Chemistry and Biochemistry of the CAS. (CC)

So far, we’ve been talking about protein databases that contain millions of sequences and hundreds of thousands of molecular structures. But plenty of other datasets exist, too – for instance, DNA sequences collected from various environments (oceans, soils, the air, and so on). With the help of AI tools, these datasets can reveal entirely new organisms, processes, reactions, and compounds that had previously escaped human notice.

There are also databases of plant DNA sequences and of fungi. The Czech project GlobalFungi, for example, contains records of fungi from more than 80,000 locations around the world. Thanks to it, scientists estimate that Earth is home to some six million species of fungi. The number of databases keeps growing. Yet already at the turn of the millennium, it became clear that rules would be needed to keep them in order and to set global standards for their control, management, maintenance, and use.

ELIXIR as a guarantee of quality
About fifteen years ago, people started talking about the so-called data lifecycle – covering everything from data planning and data collection through processing, analysis, storage, sharing, and reusing. The aim was to make sure that funding spent on research wouldn’t go to waste – for instance, by avoiding duplication of data that someone else had already created.

One of the founders of bioinformatics, Philip Bourne, estimated in 2016 that projects funded by the U.S. National Institutes of Health (NIH) had already generated 650 petabytes of data – yet only about 12 percent of it was available in NIH’s public archives. (For comparison: one petabyte could hold roughly 250 billion standard smartphone photos.) The vast majority of the data could thus be labeled “dark” or lost. In response, the NIH has spent 1.2 billion dollars in the second decade of the 21st century to support data archives and their management.

The bioinformatics community in Europe was thinking along the same lines. In 2013, it joined forces to form the ELIXIR infrastructure. One of its major contributions to global data management was the formulation of principles known by the acronym FAIR: findable, accessible, interoperable, reusable. In other words, the goal is to make scientific data easy to locate, available, capable of interconnection, and reusable in different contexts and formats.

“ELIXIR guarantees that all the data gathered in its databases carry a stamp of quality and can be trusted,” says Vondrášek, who heads the Czech node of the European infrastructure, ELIXIR CZ.

Tackling pandemics – or cancer
Artificial intelligence can be a good servant but a bad master. After all, all the men mentioned at the start of this article – the Nobel laureates behind machine learning and AI tools – have publicly and openly warned of its risks and stressed the need for ethical boundaries in its use. Vondrášek takes the same view: data scientists are well aware of the dangers of misuse in the world of big data and pay close attention to security. At the same time, he believes that the creation of AlphaFold just heralded the beginning, and that AI has much more good to offer humanity. “I believe it will help us better predict and solve challenges such as climate threats or global epidemics like COVID,” the bioinformatician says.

ELIXIR can also be described figuratively as a kind of data pipeline, streaming information from countless directions and sources. The infrastructure helps researchers navigate data, provides training, offers cloud services, and facilitates collaborative data endeavors across Europe. Its potential is enormous. With new AI tools, it should become possible to use large datasets to develop, for example, drugs for rare diseases as well as genetic or cancer-related conditions.

“We don’t have to know in advance exactly what the end result will be. What we do know for sure is that we have quality feedstock for the fast-evolving tools of artificial intelligence,” Vondrášek concludes. In the near future, we may well see more Nobel Prizes awarded in the field of big data. One thing is certain: behind the groundbreaking discoveries will stand teams whose core members include experts in digital technologies and information resources – or, facetiously put, the IT crowd.

*

The article was first published in the 2/2025 Czech issue of the quarterly A / Magazine of the CAS:

2/2025 (version for browsing)
2/2025 (version for download)

Written by: Leona Matušková, External Relations Division, CAO of the CAS
Translated by: Tereza Novická, External Relations Division, CAO of the CAS
Photo: Jana Plavec, External Relations Division, CAO of the CAS; Shutterstock

 The text and photos marked CC (and the bio profile photo) are released for use under a Creative Commons license.

Read also

• Vice-Presidents of the CAS set priorities for 2025–2029 term
• Public Hearing: “Making Sense of Open Science”
• SUNER-C concludes after three years dedicated to the renewable energy future
• A trapped state: The pandemic impact on public attitudes, trust, and behavior
• Aerial archaeology: Tracing the footsteps of our ancestors from the sky
• Archaeologists uncover ancient finds along Prague Ring Road
• Our microbiome largely depends on what we eat, says microbiologist Michal Kraus
• The ABCs of writing: Why did its invention mark a turning point for humankind?
• We learn, remember, forget… What can memory actually do? And can we outsmart it?
• New Center for Electron Microscopy in Brno opens its doors to global science