Artificial intelligence has transformed the practice of science by allowing researchers to examine the vast amounts of data generated by today’s scientific instruments. Using deep learning, you can learn from the data itself, finding a needle in a haystack of data. Artificial intelligence is driving advances in gene search, pharmacy, drug design, and compound synthesis. To extract information from new data, deep learning uses algorithms, usually neural networks trained on massive amounts of data. Following its step-by-step instructions, it is very different from traditional computing. It learns from data. Deep learning is less transparent than traditional computational programming, which leaves an important question unanswered: What does the system learn and what does it know? For fifty years, computer scientists have been trying to solve the problem of protein folding without success. In 2016, DeepMind, the artificial intelligence subsidiary of Google parent company Alphabet, launched the AlphaFold program. A protein database containing the empirically determined structures of more than 150,000 proteins was used as the training set. In less than five years, AlphaFold has solved the problem of protein folding, or at least the most important aspect of it: identifying protein structures from amino acid sequences. AlphaFold could not explain how proteins fold so quickly and precisely. This is a huge win for AI, as it has not only earned a great scientific reputation, but a major scientific breakthrough that could impact everyone’s life.
This article is reprinted from: https://www.solidot.org/story?sid=71933
This site is for inclusion only, and the copyright belongs to the original author.