{
  "docId": "019dd923-5e88-73ef-bd5d-06b04d219fea",
  "docSlug": "dd91c78f6570bf29",
  "documentTitle": "2023 Air Street Capital The State of AI Report 2023",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 60,
  "pageCount": 163,
  "prevPage": 59,
  "nextPage": 61,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 8,
  "notes": "The slide highlights a breakthrough in computational biology using language models (ESM-2) to predict protein structures.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "process_diagram",
    "bar_chart_grouped"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-06b04d219fea/60",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-06b04d219fea#slide-60",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "ESMFold structures (right) are of AF2-grade quality as measured by TM-score, which is the accuracy of the projection in comparison to the ground truth structure.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a225-f4f20fb29e1b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.18,
        "x": 0.81,
        "y": 0.4
      },
      "kind": "chart",
      "text": "Performance comparison charts (TM-score)",
      "attrs": null,
      "subkind": "bar-grouped",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9a1c8692-a77d-4c08-aa30-814d753ac1fc",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.7,
        "x": 0.22,
        "y": 0.75
      },
      "kind": "diagram",
      "text": "ESM-2 and ESMFold architecture schematic",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "3433f698-b356-49b7-bc71-10c49f44b27a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.8,
        "x": 0.02,
        "y": 0.4
      },
      "kind": "list",
      "text": "This model, Evolutionary Scale Modeling-2 (ESM-2), is used to characterize the structure of >617M metagenomic proteins (found in soil, bacteria, water, etc). ESM-2 (schematic below) offers significant speedups compared to AlphaFold-2 (AF2): these results were produced in 2 weeks using a cluster of 2,000 GPUs.\nESMFold is a fully end-to-end single-sequence structure predictor that uses a folding head for ESM-2. ESMFold structures (right) are of AF2-grade quality as measured by TM-score, which is the accuracy of the projection in comparison to the ground truth structure.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8e650014-435a-4d69-8221-e9ba72260406",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "TM-score: 0.85",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c4-7719-a225-fa7592c87280",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.95,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "Atomic-level protein structure can now be directly predicted from amino acid sequences without relying on costly and slow multiple sequence alignment (MSA). To do so, a masked language modeling objective is used over millions of evolutionarily diverse protein sequences to cause biological structure to materialize in the language model because it is linked to the sequence patterns.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "a9ad47c8-4798-4166-8b4c-2641e2a02f21",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.8,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Learning the rules of protein structure at evolutionary-scale with language models",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f0290547-4cf4-4499-ab20-9d88bd3bc188",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-41afd44ef59f",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Research + Industry sections inventory model, compute, funding facts.",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 120,
      "from": 11,
      "beatId": "019dd95a-0682-776c-8e35-523bfb7f96e6",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Escalating capabilities, compute concentration and capital flows.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 68,
      "from": 59,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-f9a066ea1e78",
      "evidence": "RFdiffusion, ESMFold, PerturbNet, AlphaMissense, Med-PaLM, PLIP, CoDoC, then meta map of AI-for-science fields.",
      "position": 7,
      "objective": "AI for science accumulates breakthroughs across bio and medicine",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 78,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}