AI in protein structure prediction with Alphafold

alphafold and ai

The power of artificial intelligence is being put to use in a revolutionary way. A team at DeepMind developed an AI system called AlphaFold which can predict with high accuracy how proteins are made up from their amino acid sequences.

What is AlphaFold?

AlphaFold is an artificial intelligence (AI) program developed by Alphabet’s/Google’s DeepMind to predict the 3D structure of proteins from their amino acid sequences with accuracy comparable to experiments. 

The system is designed as a deep learning model with input from human experts and other data sources like crystallographic structures or homology models in order to get the best possible result.

 DeepMind in partnership with EMBL-EBI (European Bioinformatics Institute) has created AlphaFold database for freely available protein predictions for scientists. The first release covered the human proteome and the proteomes of several other key organisms. In comparison, the second release contains most of the manually created UniProt entries. In 2022 they intend to extend this resource even further by adding over 100 million more proteins cataloged within  UniRef90.  

alphafold

AlphaFold AI results in competitions

 AlphaFold AI software initially has had two versions. The team of researchers using AlphaFold 1 (2018)  was ranked #1 in the overall ranking of the 13th Critical Appraisal of Methods for Protein Structure Prediction (CASP) in December 2018. The program has remarkably succeeded in predicting accurate structures with targets rated as most difficult to none existing template structures available with a partially similar sequence.

The 2020 AlphaFold 2 team was able to achieve a level of accuracy much higher than any other group. They scored above 90 for around two-thirds of proteins in the global distance test (GDT), which measures how similar your computational predictions are compared with lab experiments determined structures within an accepted cutoff used GDT’s calculations.  This means they were very close or exact matches. 

A technological breakthrough in protein structure prediction

 In 2021, the AlphaFold 2 paper was published in Nature as a pre-access publication with open source software and an interactive searchable database of protein sequences. Despite the notes that some researchers had questioned its accuracy and not revealing and solving the mechanism of protein folding. This technological breakthrough has been universally respected by researchers across disciplines for its technical achievements.

Thinking of implementing breakthrough technology in your business, contact us for a free discussion at info@datapipesoft.com