-
Categories
-
Pharmaceutical Intermediates
-
Active Pharmaceutical Ingredients
-
Food Additives
- Industrial Coatings
- Agrochemicals
- Dyes and Pigments
- Surfactant
- Flavors and Fragrances
- Chemical Reagents
- Catalyst and Auxiliary
- Natural Products
- Inorganic Chemistry
-
Organic Chemistry
-
Biochemical Engineering
- Analytical Chemistry
-
Cosmetic Ingredient
- Water Treatment Chemical
-
Pharmaceutical Intermediates
Promotion
ECHEMI Mall
Wholesale
Weekly Price
Exhibition
News
-
Trade Service
The function of a
protein is determined by its three-dimensional shape.
"In a sense, the problem has been solved," computational biologist John Moult announced
in late 2020.
London-based DeepMind has just enjoyed success in a biennial competition co-founded by Moult that tests the team's ability to predict protein structure with its revolutionary artificial intelligence (AI) tool, AlphaFold, one of
biology's toughest challenges.
Two years later, Moult's competitor, Critical Assessment of Structural Prediction (CASP), is still walking in
the long shadow of AlphaFold.
Results from this year's edition (CASP15), presented at a conference in Antalya, Turkey, this weekend, show that the most successful method for predicting protein structure from amino acid sequences includes AlphaFold, which relies on an artificial intelligence method
called deep learning.
"Everyone is using AlphaFold," says
Yang Zhang, a computational biologist at the University of Michigan in Ann Arbor.
However, AlphaFold's advances open the floodgates to new challenges in protein structure prediction — some of which are included in this year's CASP — that may require new approaches and more time to fully solve.
Mohammed AlQuraishi, a computational biologist at Columbia University in New York, said: "The low-hanging fruit has been removed
.
The next questions will be harder
.
”
CASP was founded in 1994 to improve precision in the field of protein structure prediction – advances that will accelerate the understanding of cellular components and advance drug discovery
.
During the year of the competition, the teams were tasked with using computational tools to predict the structure of proteins, which have been determined by experimental methods such as X-ray crystallography and cryo-electron microscopy, but have not yet been published
.
Entries were evaluated based on the prediction of the whole protein, or the degree to which the independent folding subunits called the domain matched the experimental structure
.
Some of AlphaFold's predictions on CASP14 are more or less indistinguishable from experimental models — the first time such accuracy
has been achieved.
Since its debut at CASP14, AlphaFold has become ubiquitous
in life sciences research.
DeepMind released the underlying code of the software in 2021 so that anyone could run the program, and this year's updated AlphaFold database contains the predicted structure (of varying mass) of almost all proteins in all organisms represented in the genomic database, totaling more than 200 million proteins
.
The success of AlphaFold and the ubiquity of the newfound present a challenge for Moult and his colleagues, who are working at the University of Maryland, Rockville, who are planning this year's CASP
.
"People say, 'Oh, we don't need CASP anymore, the problem has been solved
.
' I think this is completely wrong
.
”
At CASP15, the most successful teams were those that adapted and built AlphaFold in various ways, making modest progress
in predicting the shape of individual proteins and domains.
"The accuracy is already very high, and it's hard to improve it
," Moult says.
To make the competition more meaningful in the post-Alphafold era, Moult and his team added new challenges and tweaked
some of the existing ones.
The new tests include determining how proteins interact with other molecules, such as drugs, as well as predicting the multiple shapes
that some proteins can take.
Over the past decade, CASP has included a "complex" of multiple interacting proteins, Moult said, but accurately predicting the structure of these molecules has gained more emphasis
this year.
"It's the right thing to do," Zhang says, because predicting the structure of individual proteins or domains has largely been solved
by AlphaFold.
Arne Elofsson, a protein bioinformatician at Stockholm University, says that determining the shape of protein complexes, in particular, is an important new challenge for the field because there is still a lot of room
for improvement.
AlphaFold was originally designed to predict the shape
of individual proteins.
However, within days of its public release, other scientists said the software could be "hacked" to mimic the interaction
of multiple proteins.
In the months since, researchers have come up with countless ways to improve AlphaFold's ability to
process complexes.
To achieve this, DeepMind has even released an update
called AlphaFold-Multimer.
Such efforts seem to pay off, as the number of precision complexes of CASP15 has increased significantly compared to previous competitions, mainly due to the adoption of methods
adapted to AlphaFold.
"For us, this is a new game
that is close to experimental accuracy.
Moult said
.
"We've had some failures
as well.
"
For example, the team made surprisingly accurate predictions about a virus molecule whose function is unknown, which consists of
two identical proteins intertwined.
Ezgi Karaca, a computational structural biologist at the Center for Biomedical and Genomics in Izmir, Turkey, who evaluated the complex predictions, said the shape confused AlphaFold's previous tools
.
Karaca added that the standard version of AlphaFold failed to accurately mimic the shape of a giant 20-strept bacterial enzyme, but some teams predicted the protein's structure
by applying additional hacks to the network.
At the same time, the research team worked to predict complexes involving immune molecules called antibodies — including several antibodies attached to the SARS-CoV-2 protein — as well as related molecules
known as nanobodies.
But Karaca said there are some signs of success in some of the team's predictions, suggesting that hacking AlphaFold will help predict the shape
of these medically important molecules.
This year's CASP was also notable
for DeepMind's absence.
The company did not say why it did not attend, but released a brief statement during CASP15 congratulating the teams
that participated.
(At the same time, it also updated AlphaFold to help researchers compare
their progress to the network.
) )
Other researchers say the competition requires a considerable amount of time, and companies may feel better
spent on other challenges.
"It would be great for us if they could get involved
," Moult said.
But he added that "because these methods are so good, they can't make another big leap.
"
The researchers say that major improvements to AlphaFold will take time and may require new innovations
in machine learning and protein structure prediction.
One area under development is the application of "language models", such as those used in the Predict Text tool, to predict protein structure
.
But those methods — including one developed by social networking giant Meta — performed far less well on CASP15 than AlphaFold-based tools
.
However, these tools may help predict how mutations change protein structure—one of
several key challenges that emerge in protein structure prediction after AlphaFold's success.
Because of this, AlQuraishi said, the field is no longer focused on a single goal
.
"There are a whole bunch of these problems
.
"
