feat: proline synthesis penalty + helix bonus weight (PR #51)

Makefile

@@ -20,7 +20,7 @@ lint:
 	$(RUFF) check src tests scripts
 
 ci: lint test
-	@echo "CI passed: lint + 1158-test suite green"
+	@echo "CI passed: lint + 1168-test suite green"
 
 bench-leakage:
 	PYTHONPATH=src $(PYTHON) -m openamp_foundry.cli bench leakage \

src/openamp_foundry/scoring/activity.py

@@ -53,11 +53,15 @@ def activity_likeness_score(features: dict) -> float:
     # Normalizer 0.15 = typical_cd (0.30) × typical_mu_h (0.50).
     cross_bonus = clamp01(charge_density * mu_h / 0.15) * 0.02
 
-    # Helix propensity: Chou-Fasman Pα > 1.03 rewards helix-forming sequences
-    # Scale: indifferent (1.0) → 0.0; strong helix-former (1.20+) → 1.0
-    helix_bonus = clamp01((helix_pa - 1.0) / 0.20) * 0.01
+    # Helix propensity: Chou-Fasman Pα > 1.00 rewards helix-forming sequences.
+    # Scale: indifferent (1.0) → 0.0; strong helix-former (1.20+) → 1.0 (capped at 1.0).
+    # Weight increased 0.01 → 0.03: helical AMPs (LL-37, magainin, mastoparan) represent
+    # ~70% of membrane-active AMP families; this bonus better distinguishes them from
+    # structurally disordered or beta-sheet peptides at otherwise equal composition scores.
+    # Literature: Huang (2000) Biochim Biophys Acta; Tossi et al. (2000) Biopolymers.
+    helix_bonus = clamp01((helix_pa - 1.0) / 0.20) * 0.03
 
-    # ceiling = 0.24+0.27+0.17+0.10+0.14+0.01+0.02 = 0.95 < 1.0
+    # ceiling = 0.24+0.27+0.17+0.10+0.14+0.03+0.02 = 0.97 < 1.0
     score = (
         0.24 * length_score
         + 0.27 * charge_score

src/openamp_foundry/scoring/synthesis.py

@@ -14,13 +14,19 @@ def synthesis_feasibility_score(features: dict, valid_sequence: bool = True) ->
     - Aggregation propensity > 0: interior hydrophobic runs (VILMFW ≥ 4) and high
       beta-branched density (Val/Ile/Thr) cause on-resin aggregation and poor solubility.
       References: Quittot et al. (2017) Protein Sci; Wurth et al. (2006) J Mol Biol.
+    - Proline fraction > 15%: N-methylated backbone requires extended activation (slow couplings
+      at XP junctions); DKP formation risk at N-terminal dipeptides containing Pro during
+      piperidine Fmoc deprotection; overall coupling efficiency reduced. References: Barlos et
+      al. (1989) Int J Peptide Protein Res; Quibell et al. (1994) J Am Chem Soc; Fischer (2003)
+      Curr Opin Drug Discov Devel.
     """
     if not valid_sequence:
         return 0.0
     length = features["length"]
     repeat_run = features["longest_repeat_run"]
     cys = features["cysteine_fraction"]
     agg = features.get("aggregation_propensity", 0.0)
+    pro = features.get("proline_fraction", 0.0)
 
     score = 1.0
     if length > 30:
@@ -33,4 +39,6 @@ def synthesis_feasibility_score(features: dict, valid_sequence: bool = True) ->
         score -= 0.15
     if agg > 0.0:
         score -= min(agg * 0.25, 0.20)
+    if pro > 0.15:
+        score -= 0.10
     return round(clamp01(score), 4)

tests/test_scoring_activity.py

@@ -6,10 +6,10 @@
   - hydro_score   (weight 0.17): 1 - min(|hf - 0.45| / 0.45, 1.0)
   - aromatic_bonus (max 0.10): min(aromatic / 0.20, 1.0) * 0.10
   - amphipathicity (max 0.14): clamp01(mu_h / 0.8) * 0.14
-  - helix_bonus   (max 0.01): clamp01((helix_pa - 1.0) / 0.20) * 0.01
+  - helix_bonus   (max 0.03): clamp01((helix_pa - 1.0) / 0.20) * 0.03
   - cross_bonus   (max 0.02): clamp01(charge_density * mu_h / 0.15) * 0.02
 
-Total max = 0.24+0.27+0.17+0.10+0.14+0.01+0.02 = 0.95. Result is clamped to [0, 1] and rounded to 4 dp.
+Total max = 0.24+0.27+0.17+0.10+0.14+0.03+0.02 = 0.97. Result is clamped to [0, 1] and rounded to 4 dp.
 """
 from __future__ import annotations
 
@@ -255,6 +255,37 @@ def test_dual_property_amps_score_higher_than_single_property(self):
         assert dual > only_amph
 
 
+class TestHelixBonusWeight:
+    def test_helix_bonus_weight_0_03(self):
+        # helix_pa=1.20 → clamp01((1.20-1.0)/0.20)=1.0 → bonus=1.0*0.03=0.03
+        base = {
+            "length": 18, "charge_density": 0.30, "hydrophobic_fraction": 0.45,
+            "aromatic_fraction": 0.0, "hydrophobic_moment": 0.0,
+        }
+        s_no_helix = activity_likeness_score({**base, "helix_propensity": 1.0})
+        s_full_helix = activity_likeness_score({**base, "helix_propensity": 1.20})
+        assert abs((s_full_helix - s_no_helix) - 0.03) < 0.001
+
+    def test_helix_bonus_partial_at_pa_110(self):
+        # helix_pa=1.10 → clamp01(0.10/0.20)=0.5 → bonus=0.5*0.03=0.015
+        base = {
+            "length": 18, "charge_density": 0.30, "hydrophobic_fraction": 0.45,
+            "aromatic_fraction": 0.0, "hydrophobic_moment": 0.0,
+        }
+        s_no_helix = activity_likeness_score({**base, "helix_propensity": 1.0})
+        s_half_helix = activity_likeness_score({**base, "helix_propensity": 1.10})
+        assert abs((s_half_helix - s_no_helix) - 0.015) < 0.001
+
+    def test_score_ceiling_is_0_97(self):
+        # All components at maximum: ceiling = 0.24+0.27+0.17+0.10+0.14+0.03+0.02 = 0.97
+        all_max = {
+            "length": 18, "charge_density": 0.50, "hydrophobic_fraction": 0.45,
+            "aromatic_fraction": 0.20, "hydrophobic_moment": 0.80,
+            "helix_propensity": 1.20,
+        }
+        assert activity_likeness_score(all_max) == 0.97
+
+
 class TestKnownAMPs:
     def test_magainin_like_scores_well(self):
         # KWKLFKKIGAVLKVL — canonical AMP-like sequence

tests/test_scoring_synthesis.py

@@ -97,3 +97,49 @@ def test_missing_repeat_run_key_raises_key_error(self):
     def test_missing_cysteine_fraction_key_raises_key_error(self):
         with pytest.raises(KeyError):
             synthesis_feasibility_score({"length": 15, "longest_repeat_run": 1})
+
+
+class TestProlinePenalty:
+    def _pro_feat(self, pro: float = 0.0) -> dict:
+        return {"length": 15, "longest_repeat_run": 1, "cysteine_fraction": 0.0, "proline_fraction": pro}
+
+    def test_high_proline_penalised(self):
+        # proline_fraction > 0.15 → -0.10 penalty
+        score_low = synthesis_feasibility_score(self._pro_feat(pro=0.0))
+        score_high = synthesis_feasibility_score(self._pro_feat(pro=0.20))
+        assert score_high < score_low
+
+    def test_proline_above_threshold_penalty_value(self):
+        # 15-mer, no other defects, pro=0.20 (>0.15) → 1.0 - 0.10 = 0.90
+        score = synthesis_feasibility_score(self._pro_feat(pro=0.20))
+        assert abs(score - 0.90) < 0.001
+
+    def test_proline_at_threshold_no_penalty(self):
+        # condition is > 0.15, so exactly 0.15 must not be penalised
+        score = synthesis_feasibility_score(self._pro_feat(pro=0.15))
+        assert score == 1.0
+
+    def test_proline_just_above_threshold_penalised(self):
+        # 0.16 > 0.15 → penalty applied
+        score = synthesis_feasibility_score(self._pro_feat(pro=0.16))
+        assert abs(score - 0.90) < 0.001
+
+    def test_backward_compat_missing_proline_key(self):
+        # Features without proline_fraction (older callers) should default to 0.0 → no penalty
+        feat = {"length": 15, "longest_repeat_run": 1, "cysteine_fraction": 0.0}
+        assert synthesis_feasibility_score(feat) == 1.0
+
+    def test_proline_penalty_stacks_with_other_penalties(self):
+        # repeat_run=5 (-0.10) + cys=0.25 (-0.15) + pro=0.20 (-0.10) → 1.0 - 0.35 = 0.65
+        feat = {"length": 15, "longest_repeat_run": 5, "cysteine_fraction": 0.25, "proline_fraction": 0.20}
+        score = synthesis_feasibility_score(feat)
+        assert abs(score - 0.65) < 0.001
+
+    def test_proline_stacks_with_aggregation_propensity(self):
+        # agg=0.4: min(0.4*0.25, 0.20)=0.10 penalty; pro=0.20: 0.10 penalty → 1.0-0.20=0.80
+        feat = {
+            "length": 15, "longest_repeat_run": 1, "cysteine_fraction": 0.0,
+            "aggregation_propensity": 0.4, "proline_fraction": 0.20,
+        }
+        score = synthesis_feasibility_score(feat)
+        assert abs(score - 0.80) < 0.001