{
  "id": 3803869,
  "title": "Looking beyond natural sequences",
  "url": "https://urgent.news/2026/08/27/looking-beyond-natural-sequences",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-27T19:20:00.000Z",
  "source": {
    "name": "MIT News AI",
    "slug": "mit-news-ai",
    "url": "https://news.mit.edu/2026/looking-beyond-natural-sequences-0827"
  },
  "original_language": "en",
  "account": "Proteins are complex molecules whose function depends on their structure, which in turn is determined by the sequence of amino acids that make them up. Scientists often design new proteins using a two-step approach: first, they determine the desired structure, and then a machine-learning algorithm generates potential sequences that could adopt that structure. In nature, multiple amino acid sequences can fold into the same structure, and a single sequence may adopt different structures depending on the protein's flexibility or functional triggers. This complexity presents a challenge for AI-driven protein design, as researchers must guide the technology to recognize that many sequences can adopt the same fold.\n\nFor years, the success of protein design has been measured by whether a model can reproduce a protein sequence selected by evolution. However, Amy E. Keating, head of the Department of Biology, argues that this metric is not ideal for protein design. A new machine-learning framework called PottsMPNN, developed by Keating and her team, addresses this issue by incorporating physical principles that govern protein structure and stability, leading to improved sequence generation and the ability to predict how mutations will affect a protein's stability.\n\nPottsMPNN allows researchers to design structurally feasible proteins with sequences that do not resemble any native protein, even when designing a completely novel, designed structure. The model's success stems from its ability to better understand the sequence-energy landscape—the relationship between the identity of each amino acid and the stability of the protein. By incorporating the physical interactions between all 20 possible amino acids at each pair of positions in the protein, PottsMPNN more accurately models the sequence-energy landscape than other methods.\n\nThe researchers also introduce sets of evolutionarily related sequences into the training of the PottsMPNN framework. While this approach still relies on evolutionary information, the model demonstrates that as it depends less on native sequences, structural compatibility and energy prediction improve, even for novel proteins. Protein design in the age of AI holds immense potential for biological engineering, but it is a challenging task. Foster Birnbaum, a graduate student and lead author of the paper, expresses optimism about the future of biology, stating that the ability to design any protein we want could enable \"a potentially scary amount of biological engineering.\" Birnbaum hopes that the PottsMPNN model can be further improved and fine-tuned for specific tasks, leading to better predictions of mutations' outcomes or consequences. Ultimately, Keating emphasizes that these methods move the field toward designing useful new-to-nature proteins for diverse applications while providing a stronger foundation for future advances.",
  "summary": "A new machine-learning framework aims to improve the success rate of computational protein design while moving away from results that reproduce sequences found in nature.",
  "key_points": [
    "Amy E. Keating argues new metric for protein design success",
    "PottsMPNN model incorporates physical principles for better sequence generation",
    "Model predicts mutation effects on protein stability, enabling novel protein design"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "MIT News Research",
        "title": "Looking beyond natural sequences",
        "url": "https://urgent.news/2026/08/27/looking-beyond-natural-sequences-3806269",
        "published": "2026-08-27T19:20:00.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}