Wire EBI+ map score into titration + pca_optimization aggregate step

src/ops_model/post_process/combination/pca_optimization/__init__.py

@@ -177,6 +177,7 @@
     _compute_and_plot_embeddings,
     _score_consistency,
     _score_distinctiveness,
+    _score_ebi_plus,
 )
 from ops_model.post_process.combination.pca_optimization.phase1 import (
     pca_sweep_pooled_signal,

src/ops_model/post_process/combination/pca_optimization/aggregation.py

@@ -180,9 +180,11 @@ def _score_single_reporter_metrics(
             "distinctiveness",
             "corum",
             "chad",
+            "ebi_plus",
             "distinctiveness_all",
             "corum_all",
             "chad_all",
+            "ebi_plus_all",
         )
     }
     # Per-gene / per-complex DataFrames (raw mAP scores per entity for
@@ -194,12 +196,14 @@ def _score_single_reporter_metrics(
     result["distinctiveness_df"] = None
     result["corum_df"] = None
     result["chad_df"] = None
+    result["ebi_plus_df"] = None
     try:
         from ops_utils.analysis.map_scores import (
             phenotypic_activity_assesment,
             phenotypic_distinctivness,
             phenotypic_consistency_corum,
             phenotypic_consistency_manual_annotation,
+            phenotypic_ebi_plus,
         )
 
         g_norm = normalize_guide_adata(g_raw.copy(), norm_method)
@@ -225,6 +229,16 @@ def _score_single_reporter_metrics(
         result["distinctiveness_all"] = result["distinctiveness"]
         result["distinctiveness_df"] = dist_df
 
+        ebi_plus_df, ebi_plus_ratio = phenotypic_ebi_plus(
+            g_norm,
+            plot_results=False,
+            null_size=null_size,
+            distance=distance,
+        )
+        result["ebi_plus"] = float(ebi_plus_ratio)
+        result["ebi_plus_all"] = result["ebi_plus"]
+        result["ebi_plus_df"] = ebi_plus_df
+
         e_norm = aggregate_to_level(
             g_norm, "gene", preserve_batch_info=False, subsample_controls=False
         )
@@ -315,6 +329,7 @@ def _load_per_unit_blocks(per_unit_dir, norm_method, _logger, distance="cosine")
             "distinctiveness": reporter_metrics.get("distinctiveness_df"),
             "corum": reporter_metrics.get("corum_df"),
             "chad": reporter_metrics.get("chad_df"),
+            "ebi_plus": reporter_metrics.get("ebi_plus_df"),
         }
 
         report_rows.append(
@@ -332,14 +347,17 @@ def _load_per_unit_blocks(per_unit_dir, norm_method, _logger, distance="cosine")
                 "distinctiveness": reporter_metrics["distinctiveness"],
                 "corum": reporter_metrics["corum"],
                 "chad": reporter_metrics["chad"],
+                "ebi_plus": reporter_metrics["ebi_plus"],
                 "distinctiveness_all": reporter_metrics["distinctiveness_all"],
                 "corum_all": reporter_metrics["corum_all"],
                 "chad_all": reporter_metrics["chad_all"],
+                "ebi_plus_all": reporter_metrics["ebi_plus_all"],
             }
         )
         _logger.info(
             f"  {sig}: {g.n_obs} guides x {g.n_vars} PCs @ {g.uns.get('pca_threshold', '?')} | "
             f"act={reporter_metrics['activity']:.1%} dist={reporter_metrics['distinctiveness']:.1%} "
+            f"ebi+={reporter_metrics['ebi_plus']:.1%} "
             f"corum={reporter_metrics['corum']:.1%} chad={reporter_metrics['chad']:.1%} | "
             f"all: dist={reporter_metrics['distinctiveness_all']:.1%} "
             f"corum={reporter_metrics['corum_all']:.1%} chad={reporter_metrics['chad_all']:.1%}"
@@ -477,6 +495,10 @@ def _pivot(metric_key, key_col, value_col, out_csv):
         "distinctiveness", "perturbation", "mean_average_precision",
         overlay_dir / "gene_reporter_distinctiveness_raw.csv",
     )
+    _pivot(
+        "ebi_plus", "ebi_group", "mean_average_precision",
+        overlay_dir / "group_reporter_ebi_plus_raw.csv",
+    )
     _pivot(
         "corum", "complex_num", "mean_average_precision",
         overlay_dir / "complex_reporter_corum_consistency.csv",

src/ops_model/post_process/combination/pca_optimization/embeddings.py

@@ -349,6 +349,68 @@ def _score_distinctiveness(
         return None, 0.0
 
 
+def _score_ebi_plus(
+    adata_guide,
+    activity_map,
+    r,
+    total_feats,
+    plots_dir,
+    metrics_dir,
+    plt,
+    _logger,
+    distance="cosine",
+    suffix="",
+):
+    """Run EBI+ scoring (guide-level, complex-grouped), save CSV and plot.
+
+    EBI+ is a guide-level metric like distinctiveness: every guide's positive
+    group is its gene's EBI complex (else the gene itself), so complex members
+    collapse into one group and non-complex genes reduce to distinctiveness.
+    Returns ``(ebi_plus_map, ebi_plus_ratio)`` or ``(None, 0.0)`` on failure.
+    Computed across all geneKOs (``suffix`` only affects output filenames).
+    """
+    label = "all geneKOs"
+    if activity_map is None:
+        return None, 0.0
+    try:
+        from ops_utils.analysis.map_scores import phenotypic_ebi_plus
+
+        _logger.info(f"Running EBI+ ({label})...")
+        ebi_plus_map, ebi_plus_ratio = phenotypic_ebi_plus(
+            adata_guide,
+            plot_results=False,
+            null_size=100_000,
+            distance=distance,
+        )
+        ebi_plus_map.to_csv(
+            metrics_dir / f"phenotypic_ebi_plus{suffix}.csv", index=False
+        )
+        _logger.info(f"  EBI+ ({label}): {ebi_plus_ratio:.1%}")
+
+        try:
+            fig, ax = plt.subplots(figsize=(8, 7))
+            plot_map_scatter(
+                ax, ebi_plus_map, f"EBI+ ({label})", ebi_plus_ratio, show_ntc=False
+            )
+            fig.suptitle(
+                f"EBI+ ({label}) — {total_feats} features",
+                fontsize=13, fontweight="bold",
+            )
+            fig.tight_layout()
+            fig.savefig(
+                plots_dir / f"map_ebi_plus{suffix}.png", dpi=150, bbox_inches="tight"
+            )
+            plt.close(fig)
+            _logger.info(f"  Saved plots/map_ebi_plus{suffix}.png")
+        except Exception as exc:
+            _logger.warning(f"  EBI+ plot failed: {exc}")
+
+        return ebi_plus_map, ebi_plus_ratio
+    except Exception as exc:
+        _logger.error(f"  EBI+ ({label}) failed: {exc}")
+        return None, 0.0
+
+
 def _score_consistency(
     adata_gene,
     activity_map,

src/ops_model/post_process/combination/pca_optimization/phase2.py

@@ -54,6 +54,7 @@
     _compute_and_plot_embeddings,
     _score_consistency,
     _score_distinctiveness,
+    _score_ebi_plus,
 )
 from ops_model.post_process.combination.pca_optimization.sweep_core import (
     _init_sweep_logger,
@@ -212,6 +213,17 @@ def aggregate_channels(
         _logger,
         distance=distance,
     )
+    ebi_plus_map, ebi_plus_ratio = _score_ebi_plus(
+        adata_guide,
+        activity_map,
+        r,
+        total_feats,
+        plots_dir,
+        metrics_dir,
+        plt,
+        _logger,
+        distance=distance,
+    )
     corum_map, corum_ratio, chad_map, chad_ratio, ebi_map, ebi_ratio = _score_consistency(
         adata_gene,
         activity_map,
@@ -302,6 +314,7 @@ def aggregate_channels(
             metric_maps={
                 "Activity": activity_map,
                 "Distinctiveness": dist_map,
+                "EBI+": ebi_plus_map,
                 "EBI": ebi_map,
                 "CHAD": chad_map,
                 "CORUM": corum_map,
@@ -949,6 +962,17 @@ def apply_second_pass_pca(
         _logger,
         distance=distance,
     )
+    ebi_plus_map, ebi_plus_ratio = _score_ebi_plus(
+        adata_guide,
+        activity_map,
+        r,
+        total_feats,
+        plots_dir,
+        metrics_dir,
+        plt,
+        _logger,
+        distance=distance,
+    )
     corum_map, corum_ratio, chad_map, chad_ratio, ebi_map, ebi_ratio = _score_consistency(
         adata_gene,
         activity_map,
@@ -1007,6 +1031,7 @@ def apply_second_pass_pca(
             metric_maps={
                 "Activity": activity_map,
                 "Distinctiveness": dist_map,
+                "EBI+": ebi_plus_map,
                 "EBI": ebi_map,
                 "CHAD": chad_map,
                 "CORUM": corum_map,

src/ops_model/post_process/combination/titration/titration.py

@@ -59,7 +59,7 @@
 DOWNSAMPLE_RATIO = 0.75  # multiply cell count by this each step
 MIN_CELLS = 5_000  # stop titrating below this
 NULL_SIZE = 10_000  # smaller null for speed (per-reporter)
-METRICS = ("activity", "distinctiveness", "corum", "chad", "ebi")
+METRICS = ("activity", "distinctiveness", "corum", "chad", "ebi", "ebi_plus")
 SCALES = ("linear", "log2", "log10")  # x-axis scale variants to save
 
 # Shared plot styling / labels (used by compare_titration_versions and below)
@@ -69,20 +69,23 @@
     "corum": "mediumpurple",
     "chad": "darkorange",
     "ebi": "crimson",
+    "ebi_plus": "deeppink",
 }
 TITRATION_RATIO_LABELS = {
     "activity": "% Active",
     "distinctiveness": "% Distinctive",
     "corum": "% CORUM consistent",
     "chad": "% CHAD consistent",
     "ebi": "% EBI consistent",
+    "ebi_plus": "% EBI+ significant",
 }
 TITRATION_MAP_LABELS = {
     "activity": "Activity mAP",
     "distinctiveness": "Distinctiveness mAP",
     "corum": "CORUM mAP",
     "chad": "CHAD mAP",
     "ebi": "EBI mAP",
+    "ebi_plus": "EBI+ mAP",
 }
 SCALE_LABEL_SHORT = {"linear": "linear", "log2": "log₂", "log10": "log₁₀"}
 
@@ -330,6 +333,7 @@ def _score_all_metrics(
         phenotypic_consistency_corum,
         phenotypic_consistency_ebi,
         phenotypic_consistency_manual_annotation,
+        phenotypic_ebi_plus,
     )
 
     result = {
@@ -343,6 +347,8 @@ def _score_all_metrics(
         "chad_map_mean": math.nan,
         "ebi_ratio": math.nan,
         "ebi_map_mean": math.nan,
+        "ebi_plus_ratio": math.nan,
+        "ebi_plus_map_mean": math.nan,
     }
 
     try:
@@ -388,6 +394,23 @@ def _score_all_metrics(
     except Exception as exc:
         _logger.warning(f"    Distinctiveness scoring failed: {exc}")
 
+    try:
+        ebi_plus_map, ebi_plus_ratio = phenotypic_ebi_plus(
+            g_copairs,
+            plot_results=False,
+            null_size=NULL_SIZE,
+            distance=distance,
+        )
+        # EBI+ groups by complex, not perturbation, so subset_targets (a set of
+        # perturbation names) does not cleanly filter it — report the full ratio.
+        result["ebi_plus_ratio"] = float(ebi_plus_ratio)
+        if ebi_plus_map is not None and "mean_average_precision" in ebi_plus_map.columns:
+            result["ebi_plus_map_mean"] = float(
+                ebi_plus_map["mean_average_precision"].mean()
+            )
+    except Exception as exc:
+        _logger.warning(f"    EBI+ scoring failed: {exc}")
+
     try:
         e_norm = aggregate_to_level(
             g_copairs, "gene", preserve_batch_info=False, subsample_controls=False
@@ -503,6 +526,7 @@ def _run_titration_points(
         "corum_ratio", "corum_map_mean",
         "chad_ratio", "chad_map_mean",
         "ebi_ratio", "ebi_map_mean",
+        "ebi_plus_ratio", "ebi_plus_map_mean",
     ]
 
     rows = []
@@ -598,6 +622,7 @@ def _run_titration_points(
             f"corum={scores['corum_ratio']:.1%}±{scores.get('corum_ratio_sem', 0):.1%} "
             f"chad={scores['chad_ratio']:.1%}±{scores.get('chad_ratio_sem', 0):.1%} "
             f"ebi={scores['ebi_ratio']:.1%}±{scores.get('ebi_ratio_sem', 0):.1%} "
+            f"ebi+={scores['ebi_plus_ratio']:.1%}±{scores.get('ebi_plus_ratio_sem', 0):.1%} "
             f"({time.time() - t_step:.0f}s)"
         )
     return pd.DataFrame(rows)