AI Winner Protein Expert Team Machine learning will drive the rapid development of biotechnology

Proteins are highly complex substances
present in all living organisms.
Vikas Nanda has spent more than 20 years studying the complexity of
proteins.
The Rutgers University scientist has long studied how the unique patterns of amino acids that make up proteins determine how they become anything from hemoglobin to collagen or anything, and the mysterious self-assembly step that follows
, in which certain specific proteins come together to form more complex protein complexes.

So when scientists wanted to conduct an experiment that would compare the predictive power of a human with a deep intuitive understanding of protein design and self-assembly with an artificial intelligence computer program, Nanda, a researcher at Rutgers University's Center for Advanced Biotechnology and Medicine (CABM
), was among the best.

Now, the most successful contest results for who or what can better predict protein sequence binding are out
.
Nanda and researchers at Argonne National Laboratory in Illinois, along with colleagues from across the country, report in the journal Nature Chemistry that the battle is evenly matched, but decisive
.
Nanda and several of his colleagues won a competition against an artificial intelligence program, and the computer program narrowly won.

"Despite our extensive expertise, AI does just as well, if not better, on some datasets, showing the great potential
of machine learning in overcoming human bias," Nanda said.

Proteins are made up
of a large number of amino acids that are connected end-to-end.
These chains fold together to form three-dimensional molecules
with complex shapes.
The precise shape of each protein, as well as the amino acids it contains, determines its function
.
Scientists are very interested in the self-assembly of proteins, because a better understanding of proteins could help them design a range of revolutionary products for medical and industrial uses, such as artificial human tissue for wounds and catalysts
for new chemical products.
Some researchers work on "protein design" with their expertise in proteins, creating sequences
that produce new proteins.
Nanda's team, for example, has designed a synthetic protein that can rapidly detect VX, a dangerous nerve agent, which could pave the way
for new biosensors and treatments.

For unknown reasons, proteins self-assemble with other proteins to form complex structures
that are important in biology.
Sometimes, proteins appear to follow a certain design, such as when
they self-assemble into a protective capsid for viruses.
Sometimes when something goes wrong, they self-assemble into deadly structural forms that have been linked to
various diseases such as Alzheimer's and sickle cell disease.
"Understanding the self-assembly of proteins is critical
to making progress in many fields, including medicine and industry.
"

In the competition, Nanda and five other colleagues were given a list of proteins and asked to predict which proteins might self-assemble
.
Their predictions were compared
with those of computer programs.
Based on their observations of protein behavior, including charge patterns and hydrophobicity, human experts used a rule of thumb to select 11 proteins
that they predicted would self-assemble.
This computer program based on an advanced machine learning system selects 9 proteins
.
Humans were right
about 6 of the 11 proteins they chose.
The computer program achieved a higher percentage, and 6 of the 9 proteins it recommended were able to self-assemble
.

Experiments have shown that human experts "prefer" certain amino acids, sometimes leading them to make wrong choices
.
In addition, the computer program correctly pinpoints the properties of proteins that do not make them obvious options for self-assembly, which opens the door to
further research.
Nanda used to be skeptical of machine learning for protein assembly, but the experience made him more open
to the technology.

"We're trying to fundamentally understand the chemistry of the interactions that lead to self-assembly, so I'm concerned that using these programs will hinder important insights
," Nanda said.
"But what I'm starting to really understand is that machine learning is just another tool, just like any other tool
.
"