Meta claims new AI will help pharmaceutical companies make new drugs

According to the Wall Street Journal, Meta AI, the company's research department, used ESMFold to generate a public database cataloging the predicted structure of 617 million proteins.

Proteins are composed of long strands of amino acids and their structures are determined by a DNA genome. In the human body, proteins are essential to the function of cells and organs.

Pharmaceutical companies use proteins to develop drugs that treat many ailments and diseases, including some types of cancer, HIV, and heart disease. Those companies have started to use AI to create new drugs for diseases, some of which currently have no cure.

Alphabet, Google's parent company, has developed its own AI computer program that can predict protein structure, through its subsidiary company DeepMind Technologie. They have developed the AI AlphaFold, which predicted the structure of 214 million proteins.

According to Meta, their AI is less accurate but faster than AlphaFold, which produces roughly a third of predictions with high confidence.

A Meta research scientist, Alexander Rives, co-authored a study on ESMFold published Thursday in Science magazine.

Rives said, "Often proteins which share similar structures have similar biological functions. And if you can have a really high resolution structure, then you can begin to think about what is the actual biochemical function of these proteins."

ESMFold uses the same large language model technology that OpenAI’s ChatGPT uses.

For it to work, ESMFold must be fed letters that represent amino acids and it fills in and hidden or blank portions to generate the predicted protein, and learns as it goes, comparing its generated proteins to those that already exist.

A computational biologist at Carnegie Mellon University, Olexandr Isayev, said, "It is a big achievement, but it relies a lot on the prior work."