You may remember Google's DeepMind from their AI system, AlphaGo, which beat the world champion Go player in 2016. DeepMind has recently achieved breakthrough success in accurately predicting protein structures, using its system, AlphaFold.
TRY IT YOURSELF
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
[ Ссылка ]
[ Ссылка ]
[ Ссылка ]
To learn more about protein folding, visit:
[ Ссылка ]
[ Ссылка ]
[ Ссылка ]
FOLLOW US
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
[ Ссылка ]
[ Ссылка ]
[ Ссылка ]
DESCRIPTION
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
Biology has had a difficult problem for around 50 years - the "protein folding problem". Proteins are part of every aspect of life, including diseases, so understanding their structures is paramount.
The problem lies in the staggering number of possible formations of their 3D structures that are formed from complex amino acid sequences. Did AI just solve this perplexing problem?
In August of 2020, we covered how proteins fold, the difficult challenge of studying them, and the ways scientists were using to predict their folds. Since then, scientists have announced a major breakthrough, using AI to help solve the folding problem.
Google's DeepMind, which beat the world champion Go player in 2016 with it's AlphaGo system, recently achieved significant success in accurately predicting protein structures, using its system, AlphaFold.
Proteins are essential to life, tiny machines that support nearly all its functions. Made up of chains of amino acids, they are large, complex molecules. A protein's function mostly depends on its unique 3D structure. Some diseases arise from errors in protein folding. So, a full understanding of proteins is monumental. In addition, it could allow more precision in drug creation.
A protein can fold in an astronomical number of ways. Add to this a huge amount of amino acid sequence possibilities, and the job is incredibly challenging!r. Cyrus Levinthal estimated that a protein has 10^300 possible conformations.
Scientists have determined actual protein structures by using techniques such as nuclear magnetic resonance and X-ray crystallography. However, these methods require multi-million dollar equipment and can take years of rigorous and elaborate work for each protein. We know of over 200 million proteins, yet we've only mapped around 170,000 of them using these experimentally determined methods.
As early as the 1980's, scientists tried computational solutions, but with poor results. Starting in 1994, Professors John Moult and Krzysztof Fidelis founded Critical Assessment of Protein Structure Prediction (CASP) to accelerate this approach. Occurring every other year, the CASP challenge has prompted various teams to advance methods of identifying protein structure from sequence.
CASP chooses recently-determined protein structures that are not published in advance.
AlphaFold competed in 2018 and did very well. But at CASP 14, it effectively "crushed it" by scoring over twice as high as the next best team. AlphaFold was trained on all 170,000 publicly available protein structures. The team expanded on their prior model, using new deep learning architectures, which enabled unparalleled accuracy.
A method known as Global Distance Test (GDT) ranks results on a scale from 0-100. A score of around 90 GDT is considered on par with results obtained from experimental methods.The AlphaFold system achieved a median score of 92.4 GDT overall across all targets. One scientist used an AlphaFold prediction to determine the structure of a bacterial protein in half an hour. He had been trying for a decade to get a solution by other methods! AlphaFold was 175,000 times faster!
DeepMind cautions that there is still much to learn about how multiple proteins form complexes, and how they interact with DNA, RNA, or small molecules. With that caveat, AlphaFold is a promising example of how AI is becoming one of humanity's most useful tools, expanding our scientific knowledge at an ever-faster pace. To learn more about protein folding and deep learning, head to the links above!
HOW CAN YOU SUPPORT US?
▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀
Science to Save the World is a project of LEAF / Lifespan.io, a 501(c)(3) nonprofit organization.
► Support us with monthly donations by becoming a Lifespan Hero: [ Ссылка ]
► Subscribe: [ Ссылка ]
► Learn more, and help us: [ Ссылка ]
#AI #proteinfolding #biology #deepmind #google #alphafold #alphago #protein #deeplearning #machinelearning #neuralnets #aminoacid #sciencetosavetheworld #ststw
Did AI Just Solve a 50 Year Old Biology Problem?
Теги
DeepMindAIArtificial IntelligenceProtein FoldingBiologyProtein Structure PredictionCASPCritical Assessment of protein Structure PredictionSciencedocumentaryproteinscomputingresearchtechnologydeepmindgoogleAlphaGoGoogleHealthNobel PrizeChristian AnfinsenGoogle deepmindalphagoalphafolddeepmind alphagodeepmind alphafoldprotein foldinggoogle alphafoldgoogle alphago