{
 "source": "Goodfire EVEE, Evo Variant Effect Explorer (Pearce et al., 2026; Evo 2 7B embeddings, Arc Institute)",
 "source_url": "https://evee.goodfire.ai",
 "paper": "https://www.biorxiv.org/content/10.64898/2026.04.10.717844v3",
 "retrieved": "2026-10-20",
 "api": "public EVEE API (as used by goodfire-ai/evee-mcp)",
 "note": "EVEE predicts pathogenicity (disease), trained on ClinVar labels. Pharmacogenomic variants are mostly labelled 'drug response' in ClinVar, so these scores are not CPIC function calls.",
 "licence": "EVEE: CC BY 4.0 (Zenodo 10.5281/zenodo.19701997)",
 "derived_from": {
  "path": "assets/data/evee-snapshot.json",
  "sha256": "0a804de35c76e2e29591a3893e3040b88a490a059c235f139656dd1e4ea24f2b"
 },
 "dropped_fields": [
  "alphamissense_reference",
  "spliceai_max_reference"
 ],
 "why_dropped": "Third-party predictor values (AlphaMissense: CC BY-NC-SA 4.0; SpliceAI: CC BY-NC 4.0) are cited, never shipped (LICENSES.md, POLICY.md). Every other field is unchanged.",
 "variants": [
  {
   "rsid": "rs4986893",
   "gene": "CYP2C19",
   "allele": "*3",
   "cpic_function": "no function",
   "known_mechanism": "Premature stop p.Trp212Ter",
   "evee_variant_id": "chr10:94780652:G:A",
   "evee_url": "https://evee.goodfire.ai/#/variant/chr10:94780652:G:A",
   "consequence": "Nonsense",
   "clinvar_significance": "drug_response",
   "clinvar_review": "practice guideline",
   "evee_pathogenicity": 0.9328,
   "evee_splice_disrupting": 0.022,
   "top_probe_disruptions": [
    {
     "annotation": "interpro_Cytochrome_P450",
     "ref": 1.0,
     "alt": 0.075
    },
    {
     "annotation": "is_ptc_region",
     "ref": 0.984,
     "alt": 0.156
    },
    {
     "annotation": "ptm_phosphorylation",
     "ref": 0.833,
     "alt": 0.031
    },
    {
     "annotation": "amino_acid_N",
     "ref": 0.875,
     "alt": 0.095
    },
    {
     "annotation": "amino_acid_E",
     "ref": 0.958,
     "alt": 0.26
    }
   ],
   "evee_interpretation": null
  },
  {
   "rsid": "rs4244285",
   "gene": "CYP2C19",
   "allele": "*2",
   "cpic_function": "no function",
   "known_mechanism": "Aberrant splice site, exon 5",
   "evee_variant_id": "chr10:94781858:G:A",
   "evee_url": "https://evee.goodfire.ai/#/variant/chr10:94781858:G:A",
   "consequence": "Synonymous",
   "clinvar_significance": "likely_benign|other",
   "clinvar_review": "criteria provided, multiple submitters, no conflicts",
   "evee_pathogenicity": 0.6023,
   "evee_splice_disrupting": 0.1083,
   "top_probe_disruptions": [
    {
     "annotation": "interpro_Cytochrome_P450",
     "ref": 0.989,
     "alt": 0.122
    },
    {
     "annotation": "amino_acid_E",
     "ref": 0.959,
     "alt": 0.147
    },
    {
     "annotation": "ptm_phosphorylation",
     "ref": 0.836,
     "alt": 0.112
    },
    {
     "annotation": "amino_acid_K",
     "ref": 0.037,
     "alt": 0.695
    },
    {
     "annotation": "amino_acid_R",
     "ref": 0.976,
     "alt": 0.36
    }
   ],
   "evee_interpretation": null
  },
  {
   "rsid": "rs12248560",
   "gene": "CYP2C19",
   "allele": "*17",
   "cpic_function": "increased function",
   "known_mechanism": "Upstream regulatory, increased transcription",
   "evee_variant_id": "chr10:94761899:C:T",
   "evee_url": "https://evee.goodfire.ai/#/variant/chr10:94761899:C:T",
   "consequence": "upstream / regulatory",
   "clinvar_significance": "drug response",
   "clinvar_review": "practice guideline",
   "evee_pathogenicity": 0.0082,
   "evee_splice_disrupting": 0.0004,
   "top_probe_disruptions": [
    {
     "annotation": "ptm_phosphorylation",
     "ref": 0.275,
     "alt": 0.582
    },
    {
     "annotation": "is_branchpoint_region",
     "ref": 0.109,
     "alt": 0.399
    },
    {
     "annotation": "region_3UTR",
     "ref": 0.472,
     "alt": 0.19
    },
    {
     "annotation": "in_chain",
     "ref": 0.295,
     "alt": 0.54
    },
    {
     "annotation": "ptm_acetylation",
     "ref": 0.373,
     "alt": 0.607
    }
   ],
   "evee_interpretation": "The model assigns 1% pathogenicity, consistent with a benign call. The disruption profile shows modest positive shifts in branchpoint region probability (+0.290 at +11bp) and intron character (+0.201), alongside a decrease in 3' UTR probability (-0.282 at -3bp), suggesting the variant sits near an exon-intron boundary where the model is picking up minor contextual ambiguity rather than a clean functional disruption. None of the deltas reach magnitudes typically associated with pathogenic splicing or coding variants, and the site-level signals are similarly modest. CYP2C19 encodes a cytochrome P450 enzyme central to drug metabolism, where pathogenic variants generally disrupt the coding sequence or canonical splice sites; this variant does not show that pattern. The low pLDDT delta (-0.134) at the variant position further suggests no meaningful structural context change is predicted."
  },
  {
   "rsid": "rs1799853",
   "gene": "CYP2C9",
   "allele": "*2",
   "cpic_function": "decreased function (activity 0.5)",
   "known_mechanism": "Missense p.Arg144Cys",
   "evee_variant_id": "chr10:94942289:C:T",
   "evee_url": "https://evee.goodfire.ai/#/variant/chr10:94942289:C:T",
   "consequence": "Missense",
   "clinvar_significance": "likely_benign|drug_response|other",
   "clinvar_review": "criteria provided, multiple submitters, no conflicts",
   "evee_pathogenicity": 0.4235,
   "evee_splice_disrupting": 0.0008,
   "top_probe_disruptions": [
    {
     "annotation": "amino_acid_R",
     "ref": 0.797,
     "alt": 0.01
    },
    {
     "annotation": "amino_acid_H",
     "ref": 0.095,
     "alt": 0.807
    },
    {
     "annotation": "amino_acid_S",
     "ref": 0.014,
     "alt": 0.643
    },
    {
     "annotation": "ptm_phosphorylation",
     "ref": 0.47,
     "alt": 0.168
    },
    {
     "annotation": "amino_acid_D",
     "ref": 0.744,
     "alt": 0.481
    }
   ],
   "evee_interpretation": "The model assigns 42% pathogenicity, placing this variant in genuinely ambiguous territory where roughly 44% of similarly scored variants are pathogenic. The dominant disruption signal is an increase in PTC (premature termination codon) region probability at the variant site, rising from 0.71 to 0.92 (delta +0.21), which suggests the model interprets the local sequence context as resembling a nonsense-prone or NMD-sensitive region after the C>T change. The disordered region score drops modestly from 0.31 to 0.20 (delta -0.11), indicating the variant slightly increases predicted local structural order, which is a secondary and somewhat contradictory signal. CYP2C9 encodes a major drug-metabolizing cytochrome P450 enzyme, and Arg144 sits in a region important for substrate binding and enzyme architecture; disruption here could reduce metabolic capacity for warfarin, NSAIDs, and other substrates. However, the disruption magnitudes are moderate rather than extreme, and the absence of large negative deltas in conservation or domain membership features tempers confidence in a clearly pathogenic call."
  },
  {
   "rsid": "rs1057910",
   "gene": "CYP2C9",
   "allele": "*3",
   "cpic_function": "no function (activity 0)",
   "known_mechanism": "Missense p.Ile359Leu",
   "evee_variant_id": "chr10:94981295:A:C",
   "evee_url": "https://evee.goodfire.ai/#/variant/chr10:94981295:A:C",
   "consequence": "Missense",
   "clinvar_significance": "drug_response|other",
   "clinvar_review": "criteria provided, multiple submitters, no conflicts",
   "evee_pathogenicity": 0.1193,
   "evee_splice_disrupting": 0.0011,
   "top_probe_disruptions": [
    {
     "annotation": "amino_acid_I",
     "ref": 0.862,
     "alt": 0.025
    },
    {
     "annotation": "amino_acid_L",
     "ref": 0.001,
     "alt": 0.771
    },
    {
     "annotation": "ptm_phosphorylation",
     "ref": 0.439,
     "alt": 0.712
    },
    {
     "annotation": "interpro_Cytochrome_P450",
     "ref": 0.546,
     "alt": 0.807
    },
    {
     "annotation": "in_domain",
     "ref": 0.315,
     "alt": 0.543
    }
   ],
   "evee_interpretation": null
  },
  {
   "rsid": "rs4149056",
   "gene": "SLCO1B1",
   "allele": "c.521T>C",
   "cpic_function": "decreased function",
   "known_mechanism": "Missense p.Val174Ala (transporter)",
   "evee_variant_id": "chr12:21178614:T:C",
   "evee_url": "https://evee.goodfire.ai/#/variant/chr12:21178614:T:C",
   "consequence": "Missense",
   "clinvar_significance": "drug_response",
   "clinvar_review": "reviewed by expert panel",
   "evee_pathogenicity": 0.0346,
   "evee_splice_disrupting": 0.0022,
   "top_probe_disruptions": [
    {
     "annotation": "amino_acid_V",
     "ref": 0.798,
     "alt": 0.357
    },
    {
     "annotation": "amino_acid_A",
     "ref": 0.188,
     "alt": 0.625
    },
    {
     "annotation": "ptm_phosphorylation",
     "ref": 0.101,
     "alt": 0.533
    },
    {
     "annotation": "amino_acid_S",
     "ref": 0.497,
     "alt": 0.101
    },
    {
     "annotation": "is_dna_transposon",
     "ref": 0.401,
     "alt": 0.773
    }
   ],
   "evee_interpretation": "The model assigns a 3% pathogenicity probability, consistent with a benign call. The disruption profile is dominated by signals at positions roughly 1.5kb upstream of the variant rather than at the variant site itself, suggesting the model is capturing regional genomic context rather than direct functional disruption at Val174. The largest deltas involve transposon and repeat element annotations, branchpoint and polypyrimidine tract features, and CDS/3'UTR boundary signals, all offset by over 1,500 bp. At the variant position itself, the transmembrane and MFS transporter domain probes show only modest positive shifts (+0.15 and +0.15), indicating the local protein structural context is largely preserved. The beta strand signal shows a delta of -0.255 at 32 bp away, which is the closest structural disruption, but this is modest and does not reach the magnitude typically associated with pathogenic calls in transporter genes."
  }
 ]
}