Add dot plots to bar plots for CASP15 docking success rates

Author: amorehead

notebooks/casp15_all_multi_ligand_relaxed_rmsd_lt2_bar_chart.png

@@ -27,6 +27,7 @@
     "import numpy as np\n",
     "import pandas as pd\n",
     "import seaborn as sns\n",
+    "from matplotlib.ticker import FuncFormatter\n",
     "\n",
     "from posebench.analysis.inference_analysis_casp import (\n",
     "    CASP_BUST_TEST_COLUMNS,\n",
@@ -304,7 +305,18 @@
     "    :param method: Method name.\n",
     "    :return: Method category.\n",
     "    \"\"\"\n",
-    "    return method_category_mapping.get(method, \"DL-based blind\")"
+    "    return method_category_mapping.get(method, \"DL-based blind\")\n",
+    "\n",
+    "\n",
+    "def percent_angstrom_formatter(x, pos):\n",
+    "    \"\"\"\n",
+    "    Format function for percent/angstrom axis.\n",
+    "\n",
+    "    :param x: Value.\n",
+    "    :param pos: Position.\n",
+    "    :return: Formatted string.\n",
+    "    \"\"\"\n",
+    "    return f\"{x:.0f}% / Å\""
    ]
   },
   {
@@ -799,17 +811,108 @@
     "            width=bar_width,\n",
     "        )\n",
     "\n",
+    "        # extract raw RMSD values for each method and condition\n",
+    "        for method_idx, method in enumerate(method_mapping.values()):\n",
+    "            # get unrelaxed RMSD values grouped by target\n",
+    "            unrelaxed_rmsd_by_target = {}\n",
+    "            relaxed_rmsd_by_target = {}\n",
+    "\n",
+    "            for repeat_index in range(1, max_num_repeats_per_method + 1):\n",
+    "                # unrelaxed data\n",
+    "                casp15_unrelaxed = (\n",
+    "                    globals()[f\"scoring_results_table_{repeat_index}\"][\n",
+    "                        (\n",
+    "                            globals()[f\"scoring_results_table_{repeat_index}\"][\"dataset\"]\n",
+    "                            == \"CASP15 set\"\n",
+    "                        )\n",
+    "                        & (\n",
+    "                            globals()[f\"scoring_results_table_{repeat_index}\"][\"post-processing\"]\n",
+    "                            == \"none\"\n",
+    "                        )\n",
+    "                        & (globals()[f\"scoring_results_table_{repeat_index}\"][\"method\"] == method)\n",
+    "                    ]\n",
+    "                    .groupby(\"target\")\n",
+    "                    .agg({\"rmsd\": \"mean\"})\n",
+    "                )\n",
+    "\n",
+    "                # relaxed data\n",
+    "                casp15_relaxed = (\n",
+    "                    globals()[f\"scoring_results_table_{repeat_index}\"][\n",
+    "                        (\n",
+    "                            globals()[f\"scoring_results_table_{repeat_index}\"][\"dataset\"]\n",
+    "                            == \"CASP15 set\"\n",
+    "                        )\n",
+    "                        & (\n",
+    "                            globals()[f\"scoring_results_table_{repeat_index}\"][\"post-processing\"]\n",
+    "                            == \"energy minimization\"\n",
+    "                        )\n",
+    "                        & (globals()[f\"scoring_results_table_{repeat_index}\"][\"method\"] == method)\n",
+    "                    ]\n",
+    "                    .groupby(\"target\")\n",
+    "                    .agg({\"rmsd\": \"mean\"})\n",
+    "                )\n",
+    "\n",
+    "                # accumulate values by target\n",
+    "                for target, rmsd_value in casp15_unrelaxed.iterrows():\n",
+    "                    if target not in unrelaxed_rmsd_by_target:\n",
+    "                        unrelaxed_rmsd_by_target[target] = []\n",
+    "                    unrelaxed_rmsd_by_target[target].append(rmsd_value[\"rmsd\"])\n",
+    "\n",
+    "                for target, rmsd_value in casp15_relaxed.iterrows():\n",
+    "                    if target not in relaxed_rmsd_by_target:\n",
+    "                        relaxed_rmsd_by_target[target] = []\n",
+    "                    relaxed_rmsd_by_target[target].append(rmsd_value[\"rmsd\"])\n",
+    "\n",
+    "            # calculate average RMSD across repeats for each target\n",
+    "            unrelaxed_rmsd_averages = [\n",
+    "                np.mean(values) for values in unrelaxed_rmsd_by_target.values()\n",
+    "            ]\n",
+    "            relaxed_rmsd_averages = [np.mean(values) for values in relaxed_rmsd_by_target.values()]\n",
+    "\n",
+    "            # overlay unrelaxed RMSD points (averaged per target)\n",
+    "            if len(unrelaxed_rmsd_averages) > 0:\n",
+    "                # add small random jitter for better visibility when points overlap\n",
+    "                x_positions = np.random.normal(r1[method_idx], 0.05, len(unrelaxed_rmsd_averages))\n",
+    "                # clamp RMSD values at 100\n",
+    "                clamped_rmsd = [min(val, 100) for val in unrelaxed_rmsd_averages]\n",
+    "                axis.scatter(\n",
+    "                    x_positions,\n",
+    "                    clamped_rmsd,\n",
+    "                    alpha=0.6,\n",
+    "                    s=20,\n",
+    "                    color=\"darkred\",\n",
+    "                    edgecolors=\"black\",\n",
+    "                    linewidth=0.5,\n",
+    "                    zorder=10,\n",
+    "                )  # higher zorder to appear on top\n",
+    "\n",
+    "            # overlay relaxed RMSD points (averaged per target)\n",
+    "            if len(relaxed_rmsd_averages) > 0:\n",
+    "                x_positions = np.random.normal(r2[method_idx], 0.05, len(relaxed_rmsd_averages))\n",
+    "                # clamp RMSD values at 100\n",
+    "                clamped_rmsd = [min(val, 100) for val in relaxed_rmsd_averages]\n",
+    "                axis.scatter(\n",
+    "                    x_positions,\n",
+    "                    clamped_rmsd,\n",
+    "                    alpha=0.6,\n",
+    "                    s=20,\n",
+    "                    color=\"purple\",\n",
+    "                    edgecolors=\"black\",\n",
+    "                    linewidth=0.5,\n",
+    "                    zorder=10,\n",
+    "                )\n",
+    "\n",
     "        # add labels, titles, ticks, etc.\n",
     "        axis.set_xlabel(f\"{complex_type.title()}-ligand blind docking ({complex_license})\")\n",
-    "        axis.set_ylabel(\"Percentage of predictions\")\n",
+    "        axis.set_ylabel(\"Percentage of predictions / RMSD (Å)\")\n",
     "        axis.set_xlim(1, 23 + 0.1)\n",
     "        axis.set_ylim(0, 125)\n",
     "\n",
     "        axis.bar_label(casp15_rmsd_lt2_bar, fmt=\"{:,.1f}\", label_type=\"center\")\n",
     "        axis.bar_label(casp15_relaxed_rmsd_lt_2_bar, fmt=\"{:,.1f}\", label_type=\"center\")\n",
     "        axis.bar_label(casp15_plif_wm_bar, fmt=\"{:,.1f}\", label_type=\"center\")\n",
     "\n",
-    "        axis.yaxis.set_major_formatter(mtick.PercentFormatter())\n",
+    "        axis.yaxis.set_major_formatter(FuncFormatter(percent_angstrom_formatter))\n",
     "\n",
     "        axis.set_yticks([0, 20, 40, 60, 80, 100])\n",
     "        axis.axhline(y=0, color=\"#EAEFF8\")\n",