Brogramo
Guest
Guest

Emerging Trend in AI: AlphaFold potential benefits, limitations, and ethical implications

Introduction

Solving the protein folding problem is an emerging trend in artificial intelligence that has benefits extending beyond the protein folding problem, but it also has limitations and possible ethical concerns.

Jumper et al., (2021) highlight that “AlphaFold greatly improves the accuracy of structure prediction by incorporating novel neural network architectures and training procedures based on the evolutionary, physical and geometric constraints of protein structures.”

How AlphaFold emerged

When AlphaGo defeated professional Go player Lee Sedol in 2016, It inspired Google DeepMind to organize a small team to begin solving the protein folding problem using artificial neural networks known as AlphaFold.

What is the protein folding problem?

The protein folding problem aims to solve how a protein’s amino acid sequence gives it its structure. According to Dill et al., (2008), the protein folding problem emerged in the 1960s when scientists began using atomic-resolution to identify protein structures. Since then, the problem has been broken down into three different parts consisting of (1) the folding code, (2) protein structure prediction, and (3) the folding process.

AplphaFold milestones

In 2018, AlphaFold takes first place in the CASP13 assessment, and the AlphaFold team further expands to revamp the project. In 2020, AlphaFold wins CASP14 by a large margin and was named the solution to the protein folding problem by CASP organizers according to DeepMind (n.d.). In 2021, DeepMind released an open-source version of AlphaFold along with documentation detailing all aspects of the system.

AlphaFold potential benefits

According to DeepMind, “AlphaFold can accurately predict 3D models of protein structures and has the potential to accelerate research in every field of biology.”

Currently, AlphaFold is being used to facilitate and better understand drug development by the Drugs for Neglected Diseases Initiative (DNDi), Centre for Enzyme Innovation (CEI), University of Colorado Boulder, and the University of California.

Rob Toews, a Forbes contributor who writes about AI, notes that “AlphaFold is the most important achievement In AI”, mainly because proteins take on important roles and do most of the work in cells.

AlphaFold disadavantage

One major disadvantage of AlphaFold2 is that it uses training data to detect a protein structure given its amino acid sequences. This means that AlphaFold’s intelligence is limited to its training data, which is typical for a deep learning algorithm. While in some applications this might not be a problem, the possible amino acid sequences in proteins are nearly infinite and no amount of training data is sufficient to account for all possible sequences.

Because a protein’s shape is dynamic, some of AlphaFold’s predictions are less accurate than traditional methods. However, this limitation will cease to exist once AlphaFold adapts reinforcement learning techniques.

It is also worth noting that AlphaGo Zero was advantageous over its predecessors because it did not rely on training data and taught itself to play the game of Go from scratch.

AlphaFold ethical concerns

Although it might seem that AlphaFold is flawless, there are some ethical concerns related to its reliability. Just like any black-box algorithm, AlphaFold raises concerns because it is not easy to determine how it reached a conclusion. Another ethical concern is related to the reliability of training data AlphaFold uses. If the training data is hacked, AlphaFold will produce undesired results, or it might produce results that seem accurate but have the potential to cause harm.

References

DeepMind, (n.d.). Timeline of a breakthrough. https://www.deepmind.com/research/highlighted-research/alphafold/timeline-of-a-breakthrough

Dill, K. A., Ozkan, S. B., Shell, M. S., & Weikl, T. R. (2008). The protein folding problem. Annual review of biophysics, 37, 289–316. https://doi.org/10.1146/annurev.biophys.37.092707.153558

Jumper, J., Evans, R., Pritzel, A. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 (2021). https://doi.org/10.1038/s41586-021-03819-2

Toews, R. (2021). AlphaFold is the most important achievement in AI—ever. https://www.forbes.com/sites/robtoews/2021/10/03/alphafold-is-the-most-important-achievement-in-ai-ever/?sh=4065fff6e0af