{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "b1934393",
   "metadata": {},
   "source": [
    "# Phenotype Profile Matching: Classic Semantic Similarity vs LLM Embeddings\n",
    "\n",
    "A core use of semantic similarity is matching a **patient's phenotype profile** to\n",
    "**disease profiles**, as tools like Exomiser and Phenomizer do. This notebook compares\n",
    "ontology-based methods with LLM-embedding methods on a small, controlled version of\n",
    "that task. All methods run through the same OAK interfaces.\n",
    "\n",
    "**Task.** We take a pool of 60 OMIM diseases annotated in HPO. For each disease we simulate\n",
    "three noisy \"patients\":\n",
    "\n",
    "- 3 of the disease's phenotypes are sampled;\n",
    "- each is replaced, with probability 0.8, by an ancestor 1 or 2 levels up (imprecise\n",
    "  annotation);\n",
    "- 4 confounding phenotypes are added, drawn from the *other* diseases in the pool.\n",
    "\n",
    "Every method then ranks all 60 diseases against each patient, and we record the rank of the\n",
    "true disease. (The noise levels were chosen so the task is neither trivial nor hopeless;\n",
    "with milder noise every method scores close to perfectly.)\n",
    "\n",
    "**Methods**\n",
    "\n",
    "| family | method | description |\n",
    "|---|---|---|\n",
    "| ontology | Jaccard BMA | best-match average of pairwise ancestor-set Jaccard |\n",
    "| ontology | Resnik BMA | best-match average of the information content of the most informative common ancestor |\n",
    "| ontology | flattened closure | Jaccard of the union of all ancestors of the two profiles (one vector per profile) |\n",
    "| embedding | cosine BMA | best-match average of pairwise cosine, for each OLS model |\n",
    "| embedding | mean vector | cosine of the averaged term vectors (one vector per profile), for each OLS model |\n",
    "\n",
    "The \"flattened\" and \"mean vector\" methods merge each profile into a single vector. That\n",
    "is what makes them easy to put in an off-the-shelf vector database, so it is worth\n",
    "knowing how much accuracy that costs."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "e6861cdf",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:30:39.532070Z",
     "iopub.status.busy": "2026-10-05T01:30:39.531803Z",
     "iopub.status.idle": "2026-10-05T01:30:41.221412Z",
     "shell.execute_reply": "2026-10-05T01:30:41.219797Z"
    }
   },
   "outputs": [],
   "source": [
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\", category=UserWarning, module=\"eutils\")\n",
    "warnings.filterwarnings(\"ignore\", category=DeprecationWarning)\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "\n",
    "# reference palette (categorical slots in fixed order; sequential blue ramp)\n",
    "BLUE, ORANGE, AQUA = \"#2a78d6\", \"#eb6834\", \"#1baf7a\"\n",
    "INK, INK2, GRID, SURFACE = \"#0b0b0b\", \"#52514e\", \"#e4e3df\", \"#fcfcfb\"\n",
    "plt.rcParams.update({\n",
    "    \"figure.facecolor\": SURFACE, \"axes.facecolor\": SURFACE, \"savefig.facecolor\": SURFACE,\n",
    "    \"axes.edgecolor\": GRID, \"axes.labelcolor\": INK2, \"axes.titlecolor\": INK,\n",
    "    \"axes.grid\": True, \"grid.color\": GRID, \"grid.linewidth\": 0.6,\n",
    "    \"axes.spines.top\": False, \"axes.spines.right\": False,\n",
    "    \"xtick.color\": INK2, \"ytick.color\": INK2, \"text.color\": INK,\n",
    "    \"font.size\": 10, \"axes.titlesize\": 11, \"figure.dpi\": 110,\n",
    "})\n",
    "SEQUENTIAL = sns.blend_palette([\"#f0efec\", \"#86b6ef\", \"#2a78d6\", \"#0d366b\"], as_cmap=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "8bf66020",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:30:41.225299Z",
     "iopub.status.busy": "2026-10-05T01:30:41.224703Z",
     "iopub.status.idle": "2026-10-05T01:30:44.259661Z",
     "shell.execute_reply": "2026-10-05T01:30:44.257917Z"
    }
   },
   "outputs": [],
   "source": [
    "import random\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from oaklib import get_adapter\n",
    "from oaklib.datamodels.vocabulary import IS_A\n",
    "from oaklib.utilities.embeddings.closure_embeddings import closure_embeddings\n",
    "from oaklib.utilities.embeddings.vector_utils import cosine_similarity_matrix, jaccard_similarity_matrix\n",
    "\n",
    "hp = get_adapter(\"sqlite:obo:hp\")\n",
    "ols = get_adapter(\"ols:hp\")\n",
    "MODELS = ols.embedding_models()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "47d4b1ee",
   "metadata": {},
   "source": [
    "## Disease profiles and simulated patients\n",
    "\n",
    "Disease annotations come from the HPO `phenotype.hpoa` file. We keep OMIM diseases with\n",
    "10 to 30 positive phenotype annotations and sample a pool of 60."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "0779a285",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:30:44.263780Z",
     "iopub.status.busy": "2026-10-05T01:30:44.263144Z",
     "iopub.status.idle": "2026-10-05T01:30:48.030321Z",
     "shell.execute_reply": "2026-10-05T01:30:48.028795Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "60 diseases, 180 patients, 930 distinct phenotype terms\n",
      "Keratosis palmoplantaris striata I\n",
      "  patient: ['Abnormal nail morphology', 'Abnormality of nail color', 'Hyperkeratosis', 'Severe global developmental delay', 'Bronchiectasis', 'Postnatal growth retardation', 'Abnormal pyramidal sign']\n"
     ]
    }
   ],
   "source": [
    "HPOA = \"https://github.com/obophenotype/human-phenotype-ontology/releases/latest/download/phenotype.hpoa\"\n",
    "hpoa = pd.read_csv(HPOA, sep=\"\\t\", comment=\"#\", dtype=str)\n",
    "hpoa = hpoa[(hpoa.aspect == \"P\") & hpoa.qualifier.isna() & hpoa.database_id.str.startswith(\"OMIM:\")]\n",
    "profiles = hpoa.groupby(\"database_id\").hpo_id.apply(lambda x: sorted(set(x)))\n",
    "disease_names = hpoa.groupby(\"database_id\").disease_name.first()\n",
    "profiles = profiles[profiles.apply(len).between(10, 30)]\n",
    "\n",
    "rng = random.Random(7)\n",
    "pool = sorted(rng.sample(list(profiles.index), 60))\n",
    "pool_terms = sorted({t for d in pool for t in profiles[d]})\n",
    "parents = {}\n",
    "for s, _, o in hp.relationships(predicates=[IS_A]):\n",
    "    parents.setdefault(s, []).append(o)\n",
    "\n",
    "def generalize(t, levels):\n",
    "    for _ in range(levels):\n",
    "        if not parents.get(t):\n",
    "            break\n",
    "        t = rng.choice(parents[t])\n",
    "    return t\n",
    "\n",
    "def noisy_patient(disease, n=3, p_imprecise=0.8, n_confounders=4):\n",
    "    q = [generalize(t, rng.choice([1, 2])) if rng.random() < p_imprecise else t\n",
    "         for t in rng.sample(profiles[disease], n)]\n",
    "    others = [t for t in pool_terms if t not in profiles[disease]]\n",
    "    return q + rng.sample(others, n_confounders)\n",
    "\n",
    "patients = [(d, noisy_patient(d)) for d in pool for _ in range(3)]\n",
    "all_terms = sorted(set(pool_terms) | {t for _, q in patients for t in q})\n",
    "print(f\"{len(pool)} diseases, {len(patients)} patients, {len(all_terms)} distinct phenotype terms\")\n",
    "example, example_patient = patients[0]\n",
    "print(disease_names[example])\n",
    "print(\"  patient:\", [hp.label(t) for t in example_patient])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "15f78a7c",
   "metadata": {},
   "source": [
    "## The OAK API, for one patient and one disease\n",
    "\n",
    "The interface returns a full best-match breakdown. Below we compute the same quantities\n",
    "in bulk with numpy, so all methods can be scored quickly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "03ea5c02",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:30:48.034333Z",
     "iopub.status.busy": "2026-10-05T01:30:48.034041Z",
     "iopub.status.idle": "2026-10-05T01:31:00.474439Z",
     "shell.execute_reply": "2026-10-05T01:31:00.472420Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "best match average: 0.652\n",
      "  Abnormal nail morphology -> Nail dystrophy (0.86)\n",
      "  Abnormality of nail color -> Yellow nails (0.79)\n",
      "  Hyperkeratosis -> Orthokeratotic hyperkeratosis (0.84)\n",
      "  Severe global developmental delay -> Nail dystrophy (0.30)\n",
      "  Bronchiectasis -> Hyperhidrosis (0.35)\n",
      "  Postnatal growth retardation -> Nail dystrophy (0.36)\n",
      "  Abnormal pyramidal sign -> Streaks of hyperkeratosis along each finger onto the palm (0.47)\n"
     ]
    }
   ],
   "source": [
    "sim = ols.embedding_termset_similarity(\n",
    "    example_patient, profiles[example], model=\"text-embedding-3-large_pca512\", labels=True\n",
    ")\n",
    "print(\"best match average:\", round(sim.average_score, 3))\n",
    "for bm in sim.subject_best_matches.values():\n",
    "    print(f\"  {bm.match_source_label} -> {bm.match_target_label} ({bm.score:.2f})\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5701581d",
   "metadata": {},
   "source": [
    "## Vectors and scores\n",
    "\n",
    "Closure vectors for all terms are built in one call, so they share a vocabulary.\n",
    "Information content (IC) for Resnik is computed from the ontology structure, so each\n",
    "dimension of the closure vector can carry its ancestor's IC."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "6dc6ed1a",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:31:00.477246Z",
     "iopub.status.busy": "2026-10-05T01:31:00.476979Z",
     "iopub.status.idle": "2026-10-05T01:31:01.720068Z",
     "shell.execute_reply": "2026-10-05T01:31:01.718701Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'harrier-oss-v1-27b_pca512': 930, 'llama-embed-nemotron-8b_pca512': 930, 'text-embedding-3-large_pca512': 930, 'text-embedding-3-small_pca512': 930}\n"
     ]
    }
   ],
   "source": [
    "ids, vocab, C = closure_embeddings(hp, all_terms)\n",
    "cix = {t: i for i, t in enumerate(ids)}\n",
    "ic = dict(hp.information_content_scores(vocab, object_closure_predicates=[IS_A]))\n",
    "IC = np.array([ic.get(v, 0.0) for v in vocab], dtype=np.float32)\n",
    "\n",
    "vectors = {}\n",
    "for model in MODELS:\n",
    "    mids, M = ols.entity_embeddings(all_terms, model=model)\n",
    "    vectors[model] = dict(zip(mids, M))\n",
    "print({m: len(v) for m, v in vectors.items()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "2bbf3fa5",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:31:01.723347Z",
     "iopub.status.busy": "2026-10-05T01:31:01.723043Z",
     "iopub.status.idle": "2026-10-05T01:31:20.966559Z",
     "shell.execute_reply": "2026-10-05T01:31:20.964582Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "118800"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def bma(S):\n",
    "    # symmetric best-match average over a (patient x disease) similarity matrix\n",
    "    return 0.5 * (S.max(axis=1).mean() + S.max(axis=0).mean())\n",
    "\n",
    "def resnik_matrix(a, b):\n",
    "    # IC of the most informative common ancestor, for every pair of terms\n",
    "    A, B = C[[cix[t] for t in a]], C[[cix[t] for t in b]]\n",
    "    return (A[:, None, :] * B[None, :, :] * IC).max(axis=2)\n",
    "\n",
    "def score_all(patient, disease):\n",
    "    p_rows, d_rows = C[[cix[t] for t in patient]], C[[cix[t] for t in disease]]\n",
    "    scores = {\n",
    "        (\"ontology\", \"Jaccard BMA\"): bma(jaccard_similarity_matrix(p_rows, d_rows)),\n",
    "        (\"ontology\", \"Resnik BMA\"): bma(resnik_matrix(patient, disease)),\n",
    "        (\"ontology\", \"flattened closure\"): jaccard_similarity_matrix(\n",
    "            p_rows.max(axis=0, keepdims=True), d_rows.max(axis=0, keepdims=True))[0, 0],\n",
    "    }\n",
    "    for model in MODELS:\n",
    "        v = vectors[model]\n",
    "        P = np.array([v[t] for t in patient if t in v])\n",
    "        D = np.array([v[t] for t in disease if t in v])\n",
    "        name = model.replace(\"_pca512\", \"\")\n",
    "        scores[(\"embedding: cosine BMA\", name)] = bma(cosine_similarity_matrix(P, D))\n",
    "        scores[(\"embedding: mean vector\", name)] = cosine_similarity_matrix(\n",
    "            P.mean(axis=0, keepdims=True), D.mean(axis=0, keepdims=True))[0, 0]\n",
    "    return scores\n",
    "\n",
    "rows = []\n",
    "for i, (true_disease, patient) in enumerate(patients):\n",
    "    for candidate in pool:\n",
    "        for (family, method), score in score_all(patient, profiles[candidate]).items():\n",
    "            rows.append((i, true_disease, candidate, family, method, score))\n",
    "scores = pd.DataFrame(rows, columns=[\"patient\", \"disease\", \"candidate\", \"family\", \"method\", \"score\"])\n",
    "len(scores)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6605935f",
   "metadata": {},
   "source": [
    "## Results\n",
    "\n",
    "For each of the 180 patients and each method, we find the rank of the true disease among\n",
    "the 60 candidates (1 is best; ties share the average rank)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "79246da2",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:31:20.971621Z",
     "iopub.status.busy": "2026-10-05T01:31:20.971225Z",
     "iopub.status.idle": "2026-10-05T01:31:21.818242Z",
     "shell.execute_reply": "2026-10-05T01:31:21.816790Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>MRR</th>\n",
       "      <th>hits_at_1</th>\n",
       "      <th>hits_at_5</th>\n",
       "      <th>median_rank</th>\n",
       "      <th>MRR 95% CI low</th>\n",
       "      <th>MRR 95% CI high</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>family</th>\n",
       "      <th>method</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">ontology</th>\n",
       "      <th>Jaccard BMA</th>\n",
       "      <td>0.673</td>\n",
       "      <td>0.500</td>\n",
       "      <td>0.900</td>\n",
       "      <td>1.5</td>\n",
       "      <td>0.623</td>\n",
       "      <td>0.719</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Resnik BMA</th>\n",
       "      <td>0.564</td>\n",
       "      <td>0.389</td>\n",
       "      <td>0.817</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0.510</td>\n",
       "      <td>0.615</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">embedding: cosine BMA</th>\n",
       "      <th>text-embedding-3-large</th>\n",
       "      <td>0.536</td>\n",
       "      <td>0.372</td>\n",
       "      <td>0.756</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0.480</td>\n",
       "      <td>0.589</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>harrier-oss-v1-27b</th>\n",
       "      <td>0.509</td>\n",
       "      <td>0.317</td>\n",
       "      <td>0.783</td>\n",
       "      <td>2.0</td>\n",
       "      <td>0.453</td>\n",
       "      <td>0.560</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>text-embedding-3-small</th>\n",
       "      <td>0.473</td>\n",
       "      <td>0.294</td>\n",
       "      <td>0.711</td>\n",
       "      <td>3.0</td>\n",
       "      <td>0.419</td>\n",
       "      <td>0.526</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ontology</th>\n",
       "      <th>flattened closure</th>\n",
       "      <td>0.456</td>\n",
       "      <td>0.289</td>\n",
       "      <td>0.661</td>\n",
       "      <td>3.0</td>\n",
       "      <td>0.403</td>\n",
       "      <td>0.512</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>embedding: mean vector</th>\n",
       "      <th>text-embedding-3-large</th>\n",
       "      <td>0.399</td>\n",
       "      <td>0.228</td>\n",
       "      <td>0.611</td>\n",
       "      <td>4.0</td>\n",
       "      <td>0.346</td>\n",
       "      <td>0.452</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>embedding: cosine BMA</th>\n",
       "      <th>llama-embed-nemotron-8b</th>\n",
       "      <td>0.380</td>\n",
       "      <td>0.222</td>\n",
       "      <td>0.556</td>\n",
       "      <td>4.5</td>\n",
       "      <td>0.330</td>\n",
       "      <td>0.431</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">embedding: mean vector</th>\n",
       "      <th>harrier-oss-v1-27b</th>\n",
       "      <td>0.367</td>\n",
       "      <td>0.183</td>\n",
       "      <td>0.606</td>\n",
       "      <td>4.0</td>\n",
       "      <td>0.317</td>\n",
       "      <td>0.417</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>llama-embed-nemotron-8b</th>\n",
       "      <td>0.309</td>\n",
       "      <td>0.167</td>\n",
       "      <td>0.456</td>\n",
       "      <td>7.0</td>\n",
       "      <td>0.260</td>\n",
       "      <td>0.357</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>text-embedding-3-small</th>\n",
       "      <td>0.301</td>\n",
       "      <td>0.117</td>\n",
       "      <td>0.528</td>\n",
       "      <td>5.0</td>\n",
       "      <td>0.256</td>\n",
       "      <td>0.346</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                  MRR  hits_at_1  hits_at_5  \\\n",
       "family                 method                                                 \n",
       "ontology               Jaccard BMA              0.673      0.500      0.900   \n",
       "                       Resnik BMA               0.564      0.389      0.817   \n",
       "embedding: cosine BMA  text-embedding-3-large   0.536      0.372      0.756   \n",
       "                       harrier-oss-v1-27b       0.509      0.317      0.783   \n",
       "                       text-embedding-3-small   0.473      0.294      0.711   \n",
       "ontology               flattened closure        0.456      0.289      0.661   \n",
       "embedding: mean vector text-embedding-3-large   0.399      0.228      0.611   \n",
       "embedding: cosine BMA  llama-embed-nemotron-8b  0.380      0.222      0.556   \n",
       "embedding: mean vector harrier-oss-v1-27b       0.367      0.183      0.606   \n",
       "                       llama-embed-nemotron-8b  0.309      0.167      0.456   \n",
       "                       text-embedding-3-small   0.301      0.117      0.528   \n",
       "\n",
       "                                                median_rank  MRR 95% CI low  \\\n",
       "family                 method                                                 \n",
       "ontology               Jaccard BMA                      1.5           0.623   \n",
       "                       Resnik BMA                       2.0           0.510   \n",
       "embedding: cosine BMA  text-embedding-3-large           2.0           0.480   \n",
       "                       harrier-oss-v1-27b               2.0           0.453   \n",
       "                       text-embedding-3-small           3.0           0.419   \n",
       "ontology               flattened closure                3.0           0.403   \n",
       "embedding: mean vector text-embedding-3-large           4.0           0.346   \n",
       "embedding: cosine BMA  llama-embed-nemotron-8b          4.5           0.330   \n",
       "embedding: mean vector harrier-oss-v1-27b               4.0           0.317   \n",
       "                       llama-embed-nemotron-8b          7.0           0.260   \n",
       "                       text-embedding-3-small           5.0           0.256   \n",
       "\n",
       "                                                MRR 95% CI high  \n",
       "family                 method                                    \n",
       "ontology               Jaccard BMA                        0.719  \n",
       "                       Resnik BMA                         0.615  \n",
       "embedding: cosine BMA  text-embedding-3-large             0.589  \n",
       "                       harrier-oss-v1-27b                 0.560  \n",
       "                       text-embedding-3-small             0.526  \n",
       "ontology               flattened closure                  0.512  \n",
       "embedding: mean vector text-embedding-3-large             0.452  \n",
       "embedding: cosine BMA  llama-embed-nemotron-8b            0.431  \n",
       "embedding: mean vector harrier-oss-v1-27b                 0.417  \n",
       "                       llama-embed-nemotron-8b            0.357  \n",
       "                       text-embedding-3-small             0.346  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores[\"rank\"] = scores.groupby([\"patient\", \"family\", \"method\"]).score.rank(ascending=False, method=\"average\")\n",
    "true_ranks = scores[scores.disease == scores.candidate]\n",
    "summary = true_ranks.groupby([\"family\", \"method\"])[\"rank\"].agg(\n",
    "    MRR=lambda r: (1 / r).mean(),\n",
    "    hits_at_1=lambda r: (r <= 1).mean(),\n",
    "    hits_at_5=lambda r: (r <= 5).mean(),\n",
    "    median_rank=\"median\",\n",
    ")\n",
    "\n",
    "# 95% bootstrap interval for MRR, resampling patients\n",
    "boot_rng = np.random.default_rng(0)\n",
    "rr = 1 / true_ranks.pivot_table(index=\"patient\", columns=[\"family\", \"method\"], values=\"rank\")\n",
    "samples = np.array([rr.iloc[boot_rng.integers(0, len(rr), len(rr))].mean().values for _ in range(2000)])\n",
    "summary[\"MRR 95% CI low\"] = pd.Series(np.percentile(samples, 2.5, axis=0), index=rr.columns)\n",
    "summary[\"MRR 95% CI high\"] = pd.Series(np.percentile(samples, 97.5, axis=0), index=rr.columns)\n",
    "summary = summary.sort_values(\"MRR\", ascending=False)\n",
    "summary.round(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "af63bb51",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2026-10-05T01:31:21.821547Z",
     "iopub.status.busy": "2026-10-05T01:31:21.821265Z",
     "iopub.status.idle": "2026-10-05T01:31:22.085535Z",
     "shell.execute_reply": "2026-10-05T01:31:22.083858Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 880x484 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "FAMILY_COLORS = {\"ontology\": BLUE, \"embedding: cosine BMA\": ORANGE, \"embedding: mean vector\": AQUA}\n",
    "s = summary.reset_index()\n",
    "fig, ax = plt.subplots(figsize=(8, 4.4))\n",
    "y = np.arange(len(s))\n",
    "ax.set_axisbelow(True)\n",
    "err = [s.MRR - s[\"MRR 95% CI low\"], s[\"MRR 95% CI high\"] - s.MRR]\n",
    "ax.barh(y, s.MRR, color=[FAMILY_COLORS[f] for f in s.family], height=0.6,\n",
    "        xerr=err, error_kw={\"ecolor\": INK2, \"elinewidth\": 1, \"capsize\": 3})\n",
    "ax.set_yticks(y, [f\"{m}\" for m in s.method])\n",
    "for yi, v, hi in zip(y, s.MRR, s[\"MRR 95% CI high\"]):\n",
    "    ax.text(hi + 0.012, yi, f\"{v:.2f}\", va=\"center\", fontsize=9, color=INK)\n",
    "ax.invert_yaxis()\n",
    "ax.set_xlim(0, 1.05)\n",
    "ax.set_xlabel(\"mean reciprocal rank of the true disease (95% bootstrap CI; higher is better)\")\n",
    "ax.grid(axis=\"y\", visible=False)\n",
    "handles = [plt.Rectangle((0, 0), 1, 1, color=c) for c in FAMILY_COLORS.values()]\n",
    "ax.legend(handles, FAMILY_COLORS.keys(), frameon=False, loc=\"lower right\")\n",
    "fig.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "64907978",
   "metadata": {},
   "source": [
    "## Discussion\n",
    "\n",
    "- **The ontology wins, but not by a mile.** Best-match average over ancestor-set Jaccard\n",
    "  is the strongest method here. The best text-embedding models (text-embedding-3-large,\n",
    "  harrier) come next, close to Resnik, even though they never see the ontology graph. They\n",
    "  do well because an imprecise annotation and its more specific counterpart usually share\n",
    "  wording (\"Abnormal nail morphology\" vs \"Nail dystrophy\").\n",
    "- **Resnik is below Jaccard** in this setup. One likely reason (not tested here): IC is\n",
    "  computed from ontology structure rather than annotation frequencies, and Resnik rewards a\n",
    "  single very specific shared ancestor. The confounders are real phenotypes of other\n",
    "  diseases, so they can supply exactly such a match.\n",
    "- **Merging a profile into one vector costs accuracy, for every kind of vector.** The\n",
    "  flattened closure is 0.22 MRR below Jaccard BMA, and the mean vector is below cosine BMA\n",
    "  for every model (by 0.07 for nemotron and 0.14 to 0.17 for the others). Single-vector profiles are what make an\n",
    "  off-the-shelf vector database usable, so this gap is the price of that convenience. It\n",
    "  is consistent with the \"sets of vectors vs merged vectors\" concern.\n",
    "- **Model choice matters.** nemotron trails the other three models, consistent with its\n",
    "  weaker results in the subsumption notebook.\n",
    "\n",
    "**Caveats.** This is a small, synthetic benchmark: 60 diseases, 3 simulated patients per\n",
    "disease, and a simple noise model whose settings strongly affect absolute scores (with\n",
    "milder noise every method is near-perfect). Overlapping intervals in the chart mean those\n",
    "methods are not distinguishable at this sample size. The benchmark is meant to show how to\n",
    "run such comparisons with OAK, not to settle which method is best. To make it more\n",
    "realistic, increase the pool size, use annotation-frequency IC, add label + definition\n",
    "embeddings via the `llm:` adapter, or use real patient profiles. Vectors are cached\n",
    "locally, so re-runs only fetch new terms from OLS."
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}