Artificial intelligence can predict 3D protein structures that may lead to cancer research
The study was recently published in the journal Nature Structural and
Alpha Fold 2
The research team incorporated a type of artificial
intelligence called Alpha Fold 2 into the study. The team trained the AI to solve the
3D structure of proteins from the amino acid sequences. Alpha Fold 2 is a
neural network created by Deep Mind, an artificial intelligence company owned
by Google. Alpha Fold 2 has been accurate in its predictions of the sequences,
leaving many researchers impressed, especially when the team presented the results at an
annual assessment contest known as the Critical Assessment of protein Structure
Prediction (CASP). The research team presented the full sets of proteins for 11
different species, including humans.
The Alpha Fold 2 currently has data from over 300,000
models. Researchers evaluated
the new structures made available and compared them to the ones currently
available. They concluded that Alpha Fold 2 had contributed an additional 25% of
high-quality protein structures not previously included in the data.
Although the key role proteins play in disease such as
cancer is known, applying AI to the information will allow for a deeper comprehension
of the function of proteins at their molecular levels. The structural data about
these proteins will help researchers understand the proteins better, and to
know what other molecules may interact within the cell. This could allow
for the creation of new medication that could interfere with the protein
functions when they are altered.
Limitations within the study
The research team did find limitations with the capabilities
of using Alpha Fold 2. The team noticed that the AI had a few problems regarding
the algorithm, which had issues with recreating protein complexes, or the
collection of proteins. Proteins normally work together collectively to get a biological
function done. However, predicting how various proteins could stick together
would be desirable, but was limited when using the algorithm.