{
  "patent_number": "US 12412637",
  "country": "US",
  "title": "Computer Model Designs Proteins by Matching Structure to Sequence",
  "original_title": "Embedding-based generative model for protein design",
  "summary": "IBM's 2025 patent describes a computer system that designs new protein sequences by first understanding a target 3D protein shape and then generating a sequence that would fold into that shape.",
  "what_it_does": "This patent details a computer method for designing new protein sequences. It uses advanced AI models called transformers. First, it takes a known protein's 3D structure and converts it into a numerical representation (a 'latent fold representation') in a 'second latent space.' Simultaneously, it takes the protein's amino acid sequence and converts it into a numerical representation (a 'latent sequence representation') in a 'first latent space.' The system then trains a decoder to learn how these two representations relate, essentially creating a bridge between protein shape and protein sequence. When given a new target 3D protein structure, the system encodes it and uses the trained decoder to generate a new amino acid sequence that should fold into that specific target structure. For example, it could be used to design a protein that performs a specific function by first defining the necessary 3D shape for that function.",
  "what_it_does_not_cover": [
    "Designing biological sequences without using a transformer model encoder",
    "Methods that do not involve generating latent sequence representations",
    "Methods that do not generate latent fold representations from 3D structures",
    "Systems that do not train a decoder to learn a joint latent space between sequence and fold representations",
    "Designing sequences for structures not represented in 3D voxels",
    "Methods that do not involve inverse folding of three-dimensional structures"
  ],
  "filed": "2021-05-11",
  "granted": "2025-09-09",
  "expires": "2041-05-11",
  "status": "active",
  "holder": "International Business Machines",
  "holder_url": "https://patentbrief.org/company/international-business-machines",
  "inventors": [
    {
      "name": "Pin-Yu Chen",
      "url": "https://patentbrief.org/inventor/pin-yu-chen"
    },
    {
      "name": "Yue Cao",
      "url": "https://patentbrief.org/inventor/yue-cao"
    },
    {
      "name": "Igor Melnyk",
      "url": "https://patentbrief.org/inventor/igor-melnyk"
    },
    {
      "name": "Enara C. Vijil",
      "url": "https://patentbrief.org/inventor/enara-c-vijil"
    },
    {
      "name": "Payel Das",
      "url": "https://patentbrief.org/inventor/payel-das"
    }
  ],
  "times_cited": 0,
  "tags": [
    "biotech",
    "pharmaceutical",
    "software",
    "ai_ml"
  ],
  "abstract": "A system and method for designing protein sequences conditioned on a specific target fold. The system is a transformer-based generative framework for modeling a complex sequence-structure relationship. To mitigate the heterogeneity between the sequence domain and the fold domain, a Fold-to-Sequence model jointly learns a sequence embedding using a transformer and a fold embedding from the density of secondary structural elements in 3D voxels. The joint sequence-fold representation through novel intra-domain and cross-domain losses with an intra-domain loss forces two semantically similar (where the proteins should have the same fold(s)) samples from the same domain to be close to each other in a latent space, while a cross-domain loss forces two semantically similar samples in different domains to be closer. In an embodiment, the Fold-to-Sequence model performs design tasks that include low resolution structures, structures with a region of missing residues, and NMR structural ensembles.",
  "url": "https://patentbrief.org/patent/us/12412637/embedding-based-generative-model-for-protein-design",
  "markdown_url": "https://patentbrief.org/patent/us/12412637/embedding-based-generative-model-for-protein-design/md",
  "google_patents_url": "https://patents.google.com/patent/US12412637",
  "relatedPatents": [
    {
      "patentNumber": "20240412809",
      "countryCode": "US",
      "title": "AI Predicts 3D Protein Shapes from Genetic Code",
      "url": "https://patentbrief.org/patent/us/20240412809/protein-structure-prediction-from-amino-acid-sequences-using-self-attention-neur"
    },
    {
      "patentNumber": "12437843",
      "countryCode": "US",
      "title": "Using AI to Predict Protein Shapes for Drug Discovery",
      "url": "https://patentbrief.org/patent/us/12437843/predicting-protein-structures-using-geometry-neural-networks-that-estimate-simil"
    },
    {
      "patentNumber": "20210280268",
      "countryCode": "US",
      "title": "How Computers Predict Protein Shapes Faster Using Smart Templates",
      "url": "https://patentbrief.org/patent/us/20210280268/protein-structure-prediction-system"
    },
    {
      "patentNumber": "11176462",
      "countryCode": "US",
      "title": "Predicting How Molecules Interact with Proteins Using Two AI Streams",
      "url": "https://patentbrief.org/patent/us/11176462/system-and-method-for-prediction-of-protein-ligand-interactions-and-their-bioact"
    },
    {
      "patentNumber": "20210174903",
      "countryCode": "US",
      "title": "How to Predict Protein Shapes Better Using Lab Tests and Computers",
      "url": "https://patentbrief.org/patent/us/20210174903/enhanced-protein-structure-prediction-using-protein-homolog-discovery-and-constr"
    }
  ]
}