Wednesday, November 24, 2021

g-f(2)679 THE BIG PICTURE OF THE DIGITAL AGE (11/24/2021), nature, Artificial intelligence powers protein-folding predictions

ULTRA-condensed knowledge

"g-f" fishing of golden knowledge (GK) of the fabulous treasure of the digital ageArtificial Intelligence, Protein-folding predictions (11/24/2021)  g-f(2)426 


EXCEPTIONAL “Full Pack GK Container”

Artificial intelligence powers protein-folding predictions

      • Deep-learning algorithms such as AlphaFold2 and RoseTTAFold can now predict a protein’s 3D shape from its linear sequence — a huge boon to structural biologists.
      • For Mohammed AlQuraishi, a systems biologist at Columbia University in New York City, these possibilities suggest a new era in structural biology, emphasizing protein function over form. “For the longest time, structural biology was so focused on the individual pieces that it elevated these beautiful ribbon diagrams to being almost like an end to themselves,” he says. “Now I think structural biology is going to earn the ‘biology’ component of its name.”
      • Rarely does scientific software spark such sensational headlines. “One of biology’s biggest mysteries ‘largely solved’ by AI”, declared the BBC. Forbes called it “the most important achievement in AI — ever”. The buzz over the November 2020 debut of AlphaFold2, Google DeepMind’s artificial-intelligence (AI) system for predicting the 3D structure of proteins, has only intensified since the tool was made freely available in July.
      • The excitement relates to the software’s potential to solve one of biology’s thorniest problems — predicting the functional, folded structure of a protein molecule from its linear amino-acid sequence, right down to the position of each atom in 3D space.

      Genioux knowledge fact condensed as an image


      Another key references

      g-f(2)23 DeepMind demonstrated that AI help us understand the world: AlphaFold can predict, in record time, the shape of proteins to within the width of an atom 

      g-f(2)377 The Big Picture of Business Artificial Intelligence (7/16/2021),, New artificial intelligence software can compute protein structures in 10 minutes

      g-f(2)388 The Big Picture of Business Artificial Intelligence (7/22/2021), NYTimes, A.I. Predicts the Shapes of Molecules to Come

      Extra-condensed knowledge

      Lessons learned, nature 

      • Finding the fold
      • Even the DeepMind team was taken aback by how well AlphaFold2 performed at CASP14. “We obviously had internal benchmarking that suggested that we were going to do very well,” says Jumper. “But at the end of the day, there was still a feeling in the back of my mind: is this really, really true?”
      • CASP14 assuaged those concerns, and the past few months have seen numerous demonstrations of the capabilities and limits of AlphaFold2. In a study published alongside the paper describing the algorithm, the DeepMind team applied AlphaFold2 to a data set comprising 98.5% of the human proteome. The algorithm uses a metric called a predicted local distance difference test (pLDDT) to indicate its confidence that a particular amino acid’s position and orientation accurately reflects its real-world structure. In this way, 36% of all residues in the proteome could be resolved with very high confidence.
      • In August, researchers led by bioinformatician Alfonso Valencia at the Barcelona Supercomputing Center in Spain independently concluded that AlphaFold boosted the proportion of amino acids in human proteins that can be accurately mapped from 31% to 50%.
      • Users generally find the software straightforward to use, although they need several terabytes of disk space to download the databases and multiple graphic processing units (GPUs) to handle the analysis. “Single-structure computations are not that bad — we run it for a couple of hours,” says bioinformatician Arne Elofsson at Stockholm University. But because of their scale and the resources required, analyses of the full complement of an organism’s proteins, or proteome, are likely to be out of reach for most academic labs for the time being.
      • For researchers who wish to test-drive the software, Steinegger and his colleagues developed ColabFold, a cloud-based system that runs both AlphaFold2 and RoseTTAFold using remote databases and computing power provided by Google. The web-based interface is relatively simple: “You can plug in your sequence and then just push a button and it predicts the structure for you,” says Steinegger. But it also allows users to tinker with settings and optimize their experiments — such as by changing the number of iterations of structure prediction.

        Condensed knowledge

        Lessons learned, nature

        • Higher education
        • Deep learning incorporates machine-learning strategies in which computational neural networks are trained to recognize and interpret patterns in data. “These models don’t try to predict the structure all in one go,” says David Baker, a computational biologist at the University of Washington in Seattle. “They’re more like a physical simulation where the models are learning how to make good moves to improve the structure.” By training these algorithms with vast amounts of annotated experimental data, they can begin identifying links between sequence and structure that inform predictions for new proteins.
        • Over the past five years, multiple teams have made headway in applying deep learning to structure prediction. The first iteration of AlphaFold won CASP13 in 2018, but its performance was nowhere near the stand-out victory seen last year. Several academic laboratories subsequently developed deep-learning-based algorithms that outperformed the first generation of AlphaFold, including the Zhang lab’s D-I-TASSER, the Baker lab’s trRosetta and RaptorX, developed by Jinbo Xu and his team at the Toyota Technological Institute in Chicago, Illinois.
        • But these algorithms were generally applied as parts of a larger software pipeline, creating the potential for error and inefficiency. “You often had different components miscommunicating or not communicating optimally with one another because they were built piecemeal,” says Mohammed AlQuraishi, a systems biologist at Columbia University in New York City. These limitations have fuelled interest in end-to-end algorithms that manage the entire process from sequence to structure. DeepMind senior research scientist John Jumper, who is based in London, says that after CASP13, his team essentially discarded the first-generation AlphaFold and began to develop such a solution — AlphaFold2.

        Some relevant characteristics of this "genioux fact"

        • Category 2: The Big Picture of the Digital Age
        • [genioux fact deduced or extracted from nature]
        • This is a “genioux fact fast solution.”
        • Tag Opportunities those travelling at high speed on GKPath
        • Type of essential knowledge of this “genioux fact”: Essential Analyzed Knowledge (EAK).
        • Type of validity of the "genioux fact". 

          • Inherited from sources + Supported by the knowledge of one or more experts.


        “genioux facts”: The online programme on MASTERING “THE BIG PICTURE OF THE DIGITAL AGE”, g-f(2)679, Fernando Machuca, November 24, 2021, Corporation.


        PhD with awarded honors in computer science in France

        Fernando is the director of "genioux facts". He is the entrepreneur, researcher and professor who has a disruptive proposal in The Digital Age to improve the world and reduce poverty + ignorance + violence. A critical piece of the solution puzzle is "genioux facts"The Innovation Value of "genioux facts" is exceptional for individuals, companies and any kind of organization.

        Key “genioux facts”

        Featured "genioux fact"

        g-f(2)2393 Unlock Your Greatness: Today's Daily Dose of g-f Golden Knowledge (May 2024)

          genioux Fact post by  Fernando Machuca  and  Claude Updated May 19, 2024 Introduction: Welcome to May 2024's edition of "Unlock Y...

        Popular genioux facts, Last 30 days