Artificial intelligence has been at the forefront of lasting change in the healthcare industry, and perhaps the biggest advancement is AlphaFold. Created by Google DeepMind, AlphaFold is a deep neural network that predicts the structure of proteins simply by analyzing their amino acid sequences. The system received worldwide recognition in 2020 for solving the “protein folding problem” and subsequently won the Nobel Prize in Chemistry in 2024.
AlphaFold Description
AlphaFold is an internationally recognized algorithm in 2020 that accurately predicts protein structures based on amino acid sequences and solves the “protein folding problem.” AlphaFold promises to accelerate drug discovery and development, improve disease modeling, and otherwise transform the healthcare industry by calculating predictions in seconds.
Unraveling the mysteries of protein structure has opened up a wide range of possibilities, from accelerating drug development to developing new treatments for diseases. Below, we’ll explain more about AlphaFold, including how it works, why it’s important, its potential benefits, and some limitations to keep in mind.
What is Alphafold?
AlphaFold is an advanced algorithm that can predict the three-dimensional structure of proteins by studying one-dimensional amino acid sequences. Specifically, it is an artificial neural network consisting of multiple layers of simulated nodes that mimic the behavior of neurons in the human brain. These nodes work together to process large data sets that can be referenced to uncover complex patterns in new data.
These features make AlphaFold ideal for solving complex amino acid sequences, making it possible to accurately predict protein structures in seconds. For peace of mind, AlphaFold also shares a confidence score with its predictions, letting users know how reliable its calculations are.
Why is protein structure important?
Knowing the structure of a protein is very important because the function of a protein is closely related to its shape. The ability to predict protein structures allows researchers to better understand how proteins function and the precise role they play in cells.
After all, proteins are the building blocks of life, allowing the formation and maintenance of cellular structures. Accidents at the genetic level during protein folding can disrupt cell development and lead to disease or death in the organism. Predicting the structure of proteins not only provides insight into how proteins function in general, but also helps research in areas such as drug discovery and genetic diseases.
Protein folding problems
Proteins are made up of 20 amino acids linked together. These amino acids can be assembled into proteins in just a few milliseconds, but processing all possible combinations of amino acids to predict the shape of a protein is incredibly complex. In fact, it would take the lifetime of the universe to complete this calculation. This problem is known as Levinthal’s paradox, named after Silas Levinthal, who discovered it in the 1960s. It also doesn’t help that little is known about the actual process by which proteins fold into their final structures.
However, biochemist Christian Anfinsen contributed an important additional paper in 1972, arguing that protein structure depends on the precise sequence of amino acids. In other words, researchers need to be able to use a protein’s amino acid sequence to predict its structure without understanding the process of protein folding. This hypothesis opens the door for AlphaFold to leverage AI to overcome the computational complexity of protein folding.
How does AlphaFold work?
To handle the huge number of amino acid combinations, AlphaFold was trained on several large publicly available datasets. These include the Protein Data Bank, which contains over 256,000 protein structures, and UniProt, which lists approximately 150 million protein sequences. Using machine learning and deep learning, AlphaFold can consult these databases to recognize new amino acid sequence patterns.
DeepMind researchers also built AlphaFold using a transformer architecture, making it specialized for sequence-to-sequence tasks. In this case, AlphaFold has a built-in transformer called Invariant Point Attend that is great for solving problems related to 3D structures.
Despite extensive training data and design, AlphaFold needs to calculate two variables to predict the structure of a protein: the distance between each pair of amino acids and the angle of the chemical bonds between them. For distance, AlphaFold generates a range of possible distances with varying probabilities, combines them to create a performance score, and applies gradient descent to narrow down the answer and gradually improve the accuracy of the prediction.
AlphaFold vs. AlphaFold 2 vs. AlphaFold 3
Google DeepMind researchers presented the original version of AlphaFold at the 13th Critical Assessment of Protein Structure Prediction (CASP) competition in 2018. It demonstrated the ability of neural networks to predict protein structure by using three neural network calculations that produced accurate predictions. Since then, AlphaFold has come a long way over the next two generations.
- Alphafold 2: In 2020, AlphaFold 2 solved a protein folding problem that had puzzled scientists for 50 years. The model upgrades its predecessor’s architecture to a neural network called Evoformer that processes multiple sequence alignments to predict the structure of a single protein.
- Alphafold 3: AlphaFold 3 goes a step further than AlphaFold 2 by replacing Evoformer with a Pairformer architecture and diffusion network to predict the structure of other molecules besides DNA, RNA, ions, and proteins.
Another key difference between AlphaFold 2 and 3 is their availability. AlphaFold 2 has an Apache 2 license and can be used for both academic and commercial purposes. AlphaFold 3’s source code also has this license, but the model’s trained weights and parameters are not open source. Instead, AlphaFold 3 is subject to terms of use that currently limit it to non-commercial contexts.
What is AlphaFold used for?
AlphaFold has revolutionized a variety of scientific fields, from drug discovery to environmental conservation.
- Pharmaceutical development: Improved knowledge about proteins will allow scientists to quickly identify which proteins are ideal drug targets and develop customized medicines.
- Disease modeling: AlphaFold has helped model proteins for heart disease, leishmaniasis, and Chagas disease, accelerating research into potential treatments.
- Pandemic response: AlphaFold contributed to efforts to predict the protein structure of the Covid-19 virus and design accurate tests.
- Genetic diversity: AlphaFold can track the relationship between DNA changes and physical characteristics, shedding light on neurodivergent diseases such as autism.
- Environmental protection: Researchers are using AlphaFold to study proteins that help bees better cope with stress and disease, with the ultimate goal of promoting genetically healthier bee populations.
Advantages of AlphaFold
Previously, researchers had spent decades and millions of dollars on equipment to decipher protein structures. AlphaFold can predict these structures in seconds, making the process cheaper and faster. DeepMind has also open-sourced the code behind AlphaFold and provided free access to a protein structure database containing more than 200 million predictions, allowing all scientists to leverage AlphaFold’s discoveries to enhance their research.
Extensive knowledge of the proteins that make up the human body has helped speed drug discovery, strengthened pandemic responses, and strengthened research on neural bifurcation. Without AlphaFold, scientists would have to rely on methods that take months or even years, severely hampering the quality of care that society expects.
AlphaFold limitations
Although AlphaFold is a major step forward in fully understanding proteins, it remains inherently flawed. It inherits missing values and errors from the training data, making it susceptible to illusions that affect accuracy and reliability.
Another drawback is that AlphaFold’s predictions are based on static snapshots of proteins, whereas proteins are actually dynamic and constantly evolving within an organism. AlphaFold cannot account for chemical modifications, mutations, and other changes that proteins undergo. We also struggle with special cases, such as regions that are inherently disordered, that is, parts of proteins that do not have a well-defined structure. As a result, its applicability remains limited.
What is Alphafold?
AlphaFold is an algorithm built by Google DeepMind that famously solved the “protein folding problem” in 2020. Trained on several large datasets, AlphaFold uses its vast knowledge of proteins to identify new amino acid sequence patterns. It then calculates the distance between each pair of amino acids and the angle of their chemical bonds to predict the structure of the protein. This breakthrough has led to advances in areas such as drug development, disease modeling, and environmental conservation, among others.
What is AlphaFold 3?
AlphaFold 3 is the latest version of the AlphaFold algorithm and is the successor to previous models that focused on predicting the structure of a single protein. By combining neural networks called pairformers with diffusion networks, AlphaFold 3 can accurately predict the structures of more complex proteins and other molecules such as DNA, RNA, and ions.
Is AlphaFold a replacement for laboratory experiments?
No, while AlphaFold’s predictions support hypotheses and accelerate research on protein structures, scientists still need to conduct experiments and test their predictions to ensure the most accurate and reliable results.
How is AlphaFold used in drug discovery?
AlphaFold has provided the scientific community with deep insights into how protein structures work. Researchers are leveraging this extensive knowledge to more precisely target specific proteins when repurposing existing drugs or designing new ones, accelerating the process of drug discovery and development.
