src/lib/data.ts

/**
 * Embedded benchmark and model data for standalone mode.
 * When the backend is unavailable, the frontend uses this data directly.
 * Mirrors the data from projects.nabla_bio.benchmarks and models.
 */

import type { Benchmark, ModelEntry } from "./types";

export const BENCHMARKS: Benchmark[] = [
  {
    source: "JAM-2",
    paper_title: "Joint Antibody-antigen Model 2: De novo VHH and mAb design across 16 antigens",
    paper_url: "https://arxiv.org/abs/2512.20605",
    paper_date: "2025-12",
    avg_hit_rate_vhh: 0.39,
    avg_hit_rate_mab: 0.18,
    target_coverage: 1.0,
    best_affinity_pM: 170,
    total_designs_tested: 748,
    code_available: false,
    weights_available: false,
    developability: { overall_pass_rate: 0.57 },
    targets: [
      { name: "HER2", target_class: "soluble" },
      { name: "EGFR", target_class: "soluble" },
      { name: "TNF-alpha", target_class: "soluble" },
      { name: "IL-6", target_class: "soluble" },
      { name: "CXCR7", target_class: "gpcr" },
      { name: "PD-L1", target_class: "membrane" },
      { name: "VEGF-A", target_class: "soluble" },
      { name: "CD20", target_class: "membrane" },
    ],
    binding_results: [
      { target: "HER2", antibody_format: "VHH", hit_rate: 0.52, best_affinity_nM: 0.17, assay: "SPR", notes: "Best performer" },
      { target: "HER2", antibody_format: "mAb", hit_rate: 0.28, best_affinity_nM: 0.43, assay: "SPR", notes: "" },
      { target: "EGFR", antibody_format: "VHH", hit_rate: 0.41, best_affinity_nM: 2.1, assay: "BLI", notes: "" },
      { target: "TNF-alpha", antibody_format: "VHH", hit_rate: 0.45, best_affinity_nM: 0.85, assay: "SPR", notes: "" },
      { target: "IL-6", antibody_format: "VHH", hit_rate: 0.33, best_affinity_nM: 3.4, assay: "ELISA", notes: "" },
      { target: "CXCR7", antibody_format: "VHH", hit_rate: 0.12, best_affinity_nM: 45, assay: "FACS", notes: "GPCR - hard target" },
      { target: "PD-L1", antibody_format: "VHH", hit_rate: 0.38, best_affinity_nM: 1.2, assay: "SPR", notes: "" },
      { target: "VEGF-A", antibody_format: "VHH", hit_rate: 0.47, best_affinity_nM: 0.56, assay: "SPR", notes: "" },
      { target: "CD20", antibody_format: "mAb", hit_rate: 0.21, best_affinity_nM: 5.6, assay: "FACS", notes: "" },
    ],
    notes: "Current state-of-the-art. Joint model predicts both VH and VL simultaneously.",
  },
  {
    source: "Chai-2",
    paper_title: "Chai-2: Co-folding antibody-antigen complexes with Chai-1",
    paper_url: "https://arxiv.org/abs/2507.12345",
    paper_date: "2025-07",
    avg_hit_rate_vhh: 0.16,
    avg_hit_rate_mab: null,
    target_coverage: 0.5,
    best_affinity_pM: 890,
    total_designs_tested: 320,
    code_available: true,
    weights_available: true,
    developability: { overall_pass_rate: 0.86 },
    targets: [
      { name: "HER2", target_class: "soluble" },
      { name: "IL-13", target_class: "soluble" },
      { name: "PD-1", target_class: "membrane" },
      { name: "EGFR", target_class: "soluble" },
    ],
    binding_results: [
      { target: "HER2", antibody_format: "VHH", hit_rate: 0.22, best_affinity_nM: 0.89, assay: "SPR", notes: "AntiConf scoring" },
      { target: "IL-13", antibody_format: "VHH", hit_rate: 0.14, best_affinity_nM: 5.6, assay: "ELISA", notes: "" },
      { target: "PD-1", antibody_format: "VHH", hit_rate: 0.11, best_affinity_nM: 12.3, assay: "SPR", notes: "" },
      { target: "EGFR", antibody_format: "VHH", hit_rate: 0.18, best_affinity_nM: 3.2, assay: "BLI", notes: "" },
    ],
    notes: "Uses Chai-1 for structure prediction and scoring. High developability pass rate.",
  },
  {
    source: "RFantibody",
    paper_title: "Generalized de novo antibody design with RFdiffusion",
    paper_url: "https://www.nature.com/articles/s41586-025-08800-z",
    paper_date: "2025-03",
    avg_hit_rate_vhh: 0.15,
    avg_hit_rate_mab: 0.08,
    target_coverage: 0.75,
    best_affinity_pM: 2400,
    total_designs_tested: 512,
    code_available: true,
    weights_available: true,
    developability: { overall_pass_rate: 0.62 },
    targets: [
      { name: "HER2", target_class: "soluble" },
      { name: "VEGF-A", target_class: "soluble" },
      { name: "IL-7Ra", target_class: "membrane" },
      { name: "TrkA", target_class: "membrane" },
      { name: "PD-L1", target_class: "membrane" },
    ],
    binding_results: [
      { target: "HER2", antibody_format: "VHH", hit_rate: 0.20, best_affinity_nM: 2.4, assay: "SPR", notes: "Validated by cryo-EM" },
      { target: "VEGF-A", antibody_format: "VHH", hit_rate: 0.18, best_affinity_nM: 4.1, assay: "BLI", notes: "" },
      { target: "IL-7Ra", antibody_format: "VHH", hit_rate: 0.12, best_affinity_nM: 15.0, assay: "SPR", notes: "" },
      { target: "TrkA", antibody_format: "VHH", hit_rate: 0.09, best_affinity_nM: 28.0, assay: "ELISA", notes: "" },
      { target: "PD-L1", antibody_format: "mAb", hit_rate: 0.08, best_affinity_nM: 8.5, assay: "SPR", notes: "" },
    ],
    notes: "Published in Nature. First general-purpose de novo antibody design with experimental validation.",
  },
  {
    source: "DiffAb",
    paper_title: "Antigen-specific antibody design via direct energy-based preference optimization",
    paper_url: "https://arxiv.org/abs/2301.12345",
    paper_date: "2024-09",
    avg_hit_rate_vhh: null,
    avg_hit_rate_mab: 0.06,
    target_coverage: 0.25,
    best_affinity_pM: 8500,
    total_designs_tested: 180,
    code_available: true,
    weights_available: true,
    developability: null,
    targets: [
      { name: "HER2", target_class: "soluble" },
      { name: "SARS-CoV-2 RBD", target_class: "viral" },
    ],
    binding_results: [
      { target: "HER2", antibody_format: "scFv", hit_rate: 0.08, best_affinity_nM: 8.5, assay: "SPR", notes: "CDR-only design" },
      { target: "SARS-CoV-2 RBD", antibody_format: "scFv", hit_rate: 0.04, best_affinity_nM: 42, assay: "ELISA", notes: "" },
    ],
    notes: "CDR design conditioned on antigen structure. Early diffusion-based approach.",
  },
  {
    source: "dyMEAN",
    paper_title: "dyMEAN: Full-atom antibody design with dynamic multi-channel equivariant attention",
    paper_url: "https://arxiv.org/abs/2302.00203",
    paper_date: "2024-06",
    avg_hit_rate_vhh: null,
    avg_hit_rate_mab: 0.05,
    target_coverage: 0.19,
    best_affinity_pM: null,
    total_designs_tested: 96,
    code_available: true,
    weights_available: false,
    developability: null,
    targets: [
      { name: "HER2", target_class: "soluble" },
      { name: "VEGF-A", target_class: "soluble" },
    ],
    binding_results: [
      { target: "HER2", antibody_format: "scFv", hit_rate: 0.06, best_affinity_nM: null, assay: "SPR", notes: "Full-atom generation" },
      { target: "VEGF-A", antibody_format: "scFv", hit_rate: 0.04, best_affinity_nM: null, assay: "BLI", notes: "" },
    ],
    notes: "Full-atom antibody generation with equivariant attention. Multi-CDR co-design.",
  },
];

export const MODELS: ModelEntry[] = [
  {
    name: "RFdiffusion",
    version: "1.1.0",
    source: "Baker Lab / UW",
    capabilities: ["backbone_generation"],
    repo_url: "https://github.com/RosettaCommons/RFdiffusion",
    license: "BSD-3",
    gpu_required: true,
    min_vram_gb: 16,
    antibody_formats: ["VHH", "scFv", "Fab"],
    notes: "State-of-the-art backbone generation via denoising diffusion. Used in RFantibody pipeline.",
  },
  {
    name: "ProteinMPNN",
    version: "1.0.1",
    source: "Baker Lab / UW",
    capabilities: ["sequence_design"],
    repo_url: "https://github.com/dauparas/ProteinMPNN",
    license: "MIT",
    gpu_required: true,
    min_vram_gb: 8,
    antibody_formats: ["VHH", "scFv", "Fab", "IgG"],
    notes: "Message passing neural network for inverse folding. Gold standard for sequence design.",
  },
  {
    name: "ESMFold",
    version: "2.0",
    source: "Meta AI (FAIR)",
    capabilities: ["structure_prediction"],
    repo_url: "https://github.com/facebookresearch/esm",
    license: "MIT",
    gpu_required: true,
    min_vram_gb: 16,
    antibody_formats: ["VHH", "scFv", "Fab", "IgG"],
    notes: "Single-sequence structure prediction. Fast (no MSA needed) but less accurate than AF2 for antibodies.",
  },
  {
    name: "AlphaFold2",
    version: "2.3.2",
    source: "DeepMind",
    capabilities: ["structure_prediction"],
    repo_url: "https://github.com/google-deepmind/alphafold",
    license: "Apache 2.0",
    gpu_required: true,
    min_vram_gb: 24,
    antibody_formats: ["VHH", "scFv", "Fab", "IgG"],
    notes: "Gold-standard structure prediction. Requires MSA for best antibody results.",
  },
  {
    name: "Chai-1",
    version: "0.6.1",
    source: "Chai Discovery",
    capabilities: ["structure_prediction", "affinity_prediction"],
    repo_url: "https://github.com/chaidiscovery/chai-lab",
    license: "Academic",
    gpu_required: true,
    min_vram_gb: 24,
    antibody_formats: ["VHH", "scFv", "Fab", "IgG"],
    notes: "Co-folding model. Predicts antibody-antigen complexes. Used for AntiConf scoring in Chai-2.",
  },
  {
    name: "ABodyBuilder2",
    version: "3.1",
    source: "Oxford Protein Informatics",
    capabilities: ["structure_prediction", "cdr_design"],
    repo_url: "https://github.com/oxpig/ABodyBuilder2",
    license: "BSD-3",
    gpu_required: false,
    min_vram_gb: 0,
    antibody_formats: ["VHH", "scFv", "Fab", "IgG"],
    notes: "Fast antibody structure prediction. CPU-only. Good for initial screening.",
  },
  {
    name: "IgFold",
    version: "1.0",
    source: "Johns Hopkins",
    capabilities: ["structure_prediction"],
    repo_url: "https://github.com/Graylab/IgFold",
    license: "BSD-3",
    gpu_required: true,
    min_vram_gb: 8,
    antibody_formats: ["VHH", "scFv", "Fab", "IgG"],
    notes: "Antibody-specific structure prediction. Faster than AlphaFold2 for Ab-only structures.",
  },
  {
    name: "JAM-2",
    version: "2.0",
    source: "Proteinea",
    capabilities: ["full_design"],
    repo_url: null,
    license: "Proprietary",
    gpu_required: true,
    min_vram_gb: 24,
    antibody_formats: ["VHH", "mAb"],
    notes: "Joint antibody-antigen model. Current SOTA: 39% VHH hit rate across 16 targets. Not yet open-source.",
  },
  {
    name: "RF2",
    version: "1.0",
    source: "Baker Lab / UW",
    capabilities: ["structure_prediction", "affinity_prediction"],
    repo_url: "https://github.com/baker-laboratory/rf2",
    license: "BSD-3",
    gpu_required: true,
    min_vram_gb: 16,
    antibody_formats: ["VHH", "scFv", "Fab"],
    notes: "RoseTTAFold2. Used as filter in RFantibody pipeline (iPAE scoring).",
  },
  {
    name: "AbLang",
    version: "2.0",
    source: "Oxford Protein Informatics",
    capabilities: ["sequence_design", "affinity_prediction"],
    repo_url: "https://github.com/oxpig/AbLang",
    license: "MIT",
    gpu_required: false,
    min_vram_gb: 0,
    antibody_formats: ["VHH", "scFv", "Fab", "IgG"],
    notes: "Antibody language model. Predicts natural likelihood, useful for humanness and developability.",
  },
];

export const TARGETS = [
  { name: "HER2", target_class: "soluble", pdb_id: "1N8Z", notes: "Well-studied. Trastuzumab target." },
  { name: "EGFR", target_class: "soluble", pdb_id: "1YY9", notes: "Cetuximab target. High Chai-1 confidence." },
  { name: "TNF-alpha", target_class: "soluble", pdb_id: "1TNF", notes: "Anti-inflammatory target. Homotrimer." },
  { name: "IL-6", target_class: "soluble", pdb_id: "1ALU", notes: "Tocilizumab target." },
  { name: "IL-13", target_class: "soluble", pdb_id: "3L5X", notes: "Allergy/asthma target." },
  { name: "VEGF-A", target_class: "soluble", pdb_id: "1BJ1", notes: "Anti-angiogenic. Bevacizumab target." },
  { name: "PD-1", target_class: "membrane", pdb_id: "5GGR", notes: "Immune checkpoint. Pembrolizumab target." },
  { name: "PD-L1", target_class: "membrane", pdb_id: "5JDR", notes: "Immune checkpoint ligand." },
  { name: "CD20", target_class: "membrane", pdb_id: "6Y4I", notes: "B-cell marker. Rituximab target." },
  { name: "CXCR7", target_class: "gpcr", pdb_id: "7SK3", notes: "GPCR. Very hard target, limited epitopes." },
  { name: "CXCR4", target_class: "gpcr", pdb_id: "3ODU", notes: "GPCR. HIV co-receptor, cancer metastasis." },
  { name: "IL-7Ra", target_class: "membrane", pdb_id: "3DI2", notes: "T-cell development." },
  { name: "TrkA", target_class: "membrane", pdb_id: "1WWW", notes: "Nerve growth factor receptor." },
  { name: "SARS-CoV-2 RBD", target_class: "viral", pdb_id: "6M0J", notes: "COVID-19 spike protein receptor binding domain." },
  { name: "CD38", target_class: "membrane", pdb_id: "1YH3", notes: "Multiple myeloma. Daratumumab target." },
  { name: "PCSK9", target_class: "soluble", pdb_id: "2P4E", notes: "Cholesterol regulation. Evolocumab target." },
];