Frequently asked questions (FAQ’s) about Alpha Fold

What is Alpha fold?

AlphaFold is a Google DeepMind-developed artificial intelligence (AI) programme that predicts protein structure. The programme is intended to be a deep learning system.

What is deep learning?

Deep learning is an area of machine learning in which algorithms are structured in layers to construct a “artificial neural network” that can learn and make intelligent decisions on its own.

What is protein structure?

The three-dimensional arrangement of atoms in an amino acid chain molecule is known as protein structure. Proteins are polymers – specifically polypeptides – formed from amino acid sequences, the polymer’s monomers. Protein structure is classified into four levels: primary, secondary, tertiary, and quaternary. To completely comprehend how a protein functions, it is essential to understand the nature and function of each level of protein structure. They are essential for the structure, function, and regulation of the body’s tissues and organs and do the majority of the work in cells. Proteins are composed of hundreds or thousands of smaller units known as amino acids that are linked together in lengthy chains. A protein structure gives us a better understanding of how a protein operates, allowing us to develop theories about how to alter, control, or modify it. Knowing the structure of a protein, for example, may allow you to create site-directed alterations with the goal of affecting function. Protein structure is determined by the amino acid sequence and local, low-energy chemical interactions formed between atoms in the polypeptide backbone and amino acid side chains. Protein structure is important for protein function; if a protein loses its form at any structural level, it may no longer be functional.

What is protein folding?

Protein folding is the physical process by which a protein chain is converted to its native three-dimensional structure, often a “folded” configuration that allows the protein to operate biologically.

Why understanding protein folding is important?

Protein folding takes place in a cellular compartment known as the endoplasmic reticulum. This is an important biological process because proteins must be folded into certain three-dimensional forms in order to function properly. Proteins that are unfolded or misfolded contribute to the pathophysiology of many diseases.

Is Alpha fold 2 available?

Although the source code for AlphaFold 2 is freely available to everybody, including commercial businesses, it may not be particularly beneficial for researchers lacking technical skills.

How accurate is AlphaFold?

It is said to be roughly 95% accurate at determining the positions of individual atoms in proteins with an accuracy of Ångstrom for the most well-studied proteins.

What does solving the protein folding problem mean?

Since it was first raised roughly 50 years ago, the ‘protein folding problem’ has been a source of concern. In a word, understanding how a protein folds is critical to understanding how it will function, which could lead to solutions to huge problems like how to cure diseases like cancer.

Why Alphafold is popular in protein structure prediction?

There have been two major versions of the AlphaFold AI software. In December 2018, a team of researchers who employed AlphaFold 1 (2018) came in first place overall in the 13th Critical Assessment of Techniques for Protein Structure Prediction (CASP). The software was especially good at predicting the most precise structure for targets classified as the most difficult by the competition organisers, where no existing template structures from proteins with a partially similar sequence were available.

In November 2020, a team that employed AlphaFold 2 (2020) placed again in the CASP competition. The team outperformed all other groups in terms of accuracy. It scored above 90 in CASP’s global distance test (GDT), a test that analyses the degree to which a computer programme predicted structure is comparable to the lab experiment determined structure, with 100 being a total match within the distance cutoff used for calculating GDT.

A score of more than 90 in CASP’s global distance test (GDT) for AlphaFold 2 is regarded as a notable success in computational biology and considerable progress toward a decades-old grand problem in biology.

The results of AlphaFold 2 at CASP were hailed as “astounding” and “transformational.” Some researchers highlighted that the accuracy is not high enough for one-third of its predictions, and that it does not reveal the mechanism or rules of protein folding, implying that the protein folding problem is not solved. Nonetheless, the technical achievement has received significant acclaim.

What is Critical Assessment of Techniques for Protein Structure Prediction (CASP)?

Since 1994, the Critical Assessment of Protein Structure Prediction (CASP) has been a community-wide, global project for protein structure prediction. CASP allows research organisations to objectively test their structure prediction methods and offers the scientific community and software users with an independent assessment of the state of the art in protein structure modelling. Although the primary purpose of CASP is to enhance methods for recognising protein three-dimensional structure from its amino acid sequence, many people regard the experiment as a “world championship” in this field of research. On a regular basis, more than 100 research groups from across the world participate in CASP, and it is not uncommon for entire groups to cease their other research for months while they focus on getting their servers ready for the experiment and completing the extensive forecasts.

What is AlphaFold Protein Structure Database?

The AlphaFold Protein Structure Database, a collaboration between AlphaFold and EMBL-EBI, was unveiled on July 22, 2021. The database contains AlphaFold-predicted structures of virtually the entire UniProt proteome of humans and 20 model species at the time of debut. AlphaFold intends to expand the library with new sequences. UniProt-KB and InterProt have been updated as of July 2021 to provide AlphaFold predictions when they are available.

Does Alphafold solved protein folding problem?

Protein folding is a complex problem with many moving pieces. The structure prediction problem is a crucial piece: determining the dominant structure of a protein from its sequence (a parts list, if you will). Another issue is the folding mechanism: how does a protein get to its folded state? There are also a slew of related issues to consider, such as what kind of moving parts do folded proteins have. How do these dynamics allow proteins to communicate information and cargo? How can we create medications that turn proteins on (or off)? How can we create proteins that can fulfil novel functions?

It is unknown, however, how well AlphaFold 2 structural predictions will hold true for proteins bound in complexes with other proteins and substances. This was neither a condition of the CASP competition in which AlphaFold competed, nor was it something that AlphaFold had planned for. Where AlphaFold 2 did predict structures for proteins that had strong interactions with other copies of themselves or with other structures, the predictions tended to be the least comprehensive and accurate. Given that many of the most important biological machines in a cell are made up of such complexes or entail the alteration of protein structures in the presence of other molecules, this is a subject that will continue to pique the interest of researchers.

With so little known about the internal patterns employed by AlphaFold 2 to create its predictions, it is unknown how much the program’s ability to detect novel folds may be hampered if such folds are not well represented in known protein structures in structure databases. Furthermore, it is unknown to what extent protein structures in such databases, which are primarily made up of proteins crystallised to X-ray resolution, are representative of ordinary proteins that have not yet been crystallised. Furthermore, how indicative frozen protein crystal structures are of the dynamic configurations found in living cells is uncertain. AlphaFold 2’s troubles with protein NMR structures may not be a positive indicator.

While the resolution of AlphaFold 2’s structures is excellent, the accuracy with which binding sites are modelled must be even higher: typically, molecular docking studies require atomic positions to be accurate within a 0.3 margin, but the predicted protein structure has an RMSD of no more than 0.9 for all atoms. As a result, the structures in AlphaFold 2 may be of little use in such scenarios. Furthermore, because prediction of small-molecule binding is still not very good, computational prediction of drug targets is simply not ready to take over as the “backbone” of corporate drug discovery—and thus “protein structure determination is not a rate-limiting step in drug discovery in general,” as Science columnist Derek Lowe points out. [63] Furthermore, even after establishing the structure of a protein, comprehending how it operates, what it does, and how this relates to larger biological processes can be extremely difficult. [64] Nonetheless, if a greater understanding of protein structure leads to a better understanding of individual disease mechanisms and, ultimately, better medication targets, or if a better understanding of the distinctions between human and animal models leads to improvements, this could lead to advancements.
Finally, some have noticed that even a perfect solution to the protein prediction problem would leave unresolved questions about protein folding—that is, completely understanding how the folding process occurs in nature (and how sometimes they can also misfold).

While AlphaFold hasn’t solved protein folding, it should be an enabling technology. For example, we could use it to predict the dominant structures of proteins whose structures haven’t been determined experimentally.Despite these flaws, AlphaFold 2 was lauded as a great technological and intellectual accomplishment.

References:
1.”AlphaFold: Using AI for scientific discovery”. Deepmind. Retrieved 30 November 2020.
2.Robert F. Service, ‘The game has changed.’ AI triumphs at solving protein structures, Science, 30 November 2020
3.DeepMind’s protein-folding AI has solved a 50-year-old grand challenge of biology”. MIT Technology Review. Retrieved 30 November 2020.
4.Kryshtafovych, A; Monastyrskyy, B; Fidelis, K (2014). “CASP prediction center infrastructure and evaluation measures in CASP10 and CASP ROLL”. Proteins: Structure, Function, and Bioinformatics. 82 Suppl 2: 7–13. doi:10.1002/prot.24399. PMC 4396618. PMID 24038551.
5.Greg Bowman, Protein folding and related problems remain unsolved despite AlphaFold’s advance, Folding@home blog, 8 December 2020