move affinity initialization to optimization function

meyer-lab · Dec 18, 2024 · c3ee9aa · c3ee9aa
1 parent 9b54d81
commit c3ee9aa
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 17 deletions.
diff --git a/bicytok/figures/figure1.py b/bicytok/figures/figure1.py
@@ -63,7 +63,6 @@ def makeFigure():
     )
     targCell = "Treg"
     offTargCells = cells[cells != targCell]
-    startingAff = 8.0
 
     epitopesList = pd.read_csv(
         path_here / "bicytok" / "data" / "epitopeList.csv"
@@ -76,12 +75,10 @@ def makeFigure():
 
     for targetPairs in allTargets:
         print(targetPairs)
-        prevOptAffs = [startingAff]
         valencies = [signal[1]]
         targets = []
         naming = []
         for target, valency in targetPairs:
-            prevOptAffs.append(8.0)
             targets.append(target)
             valencies.append(valency)
             naming.append(f"{target} ({valency})")
@@ -99,7 +96,6 @@ def makeFigure():
 
         for dose in doseVec:
             optSelec, optParams = optimizeSelectivityAffs(
-                initialAffs = prevOptAffs,
                 targRecs = targRecs.to_numpy(),
                 offTargRecs = offTargRecs.to_numpy(),
                 dose = dose,

diff --git a/bicytok/figures/figure4.py b/bicytok/figures/figure4.py
@@ -60,11 +60,9 @@ def makeFigure():
                 # targeting SIGNAL and 2 subunits corresponding to target1?
                 # Right now it's just 1 subunit for target1.
                 targetsBoth = [target1]
-                optAffs = [8.0, 8.0]
                 valenciesBoth = np.array([[signal[1], valencies[i]]])
             else:
                 targetsBoth = [target1, target2]
-                optAffs = [8.0, 8.0, 8.0]
                 valenciesBoth = np.array([[signal[1], valencies[i], valencies[j]]])
 
             dfTargCell = epitopesDF.loc[
@@ -77,7 +75,6 @@ def makeFigure():
             offTargRecs = dfOffTargCell[[signal[0]] + targetsBoth]
 
             optSelec, optParams = optimizeSelectivityAffs(
-                initialAffs = optAffs,
                 targRecs = targRecs.to_numpy(),
                 offTargRecs = offTargRecs.to_numpy(),
                 dose = dose,

diff --git a/bicytok/figures/figure5.py b/bicytok/figures/figure5.py
@@ -93,7 +93,6 @@ def makeFigure():
     )
 
     for valency in valencies:
-        prevOptAffs = [8.0, 8.0, 8.0]
         for targets in allTargets:
             modelValencies = np.array([[signal_valency, valency, valency]])
 
@@ -107,13 +106,11 @@ def makeFigure():
             offTargRecs = dfOffTargCell[[signal_receptor] + targets]
 
             optSelec, optParams = optimizeSelectivityAffs(
-                initialAffs = prevOptAffs,
                 targRecs = targRecs,
                 offTargRecs = offTargRecs,
                 dose = dose,
                 valencies = modelValencies
             )
-            prevOptAffs = optParams
             select = (1 / optSelec,)
 
             non_marker_columns = ["CellType1", "CellType2", "CellType3", "Cell"]

diff --git a/bicytok/selectivityFuncs.py b/bicytok/selectivityFuncs.py
@@ -59,6 +59,7 @@ def minOffTargSelec(
     """
 
     # Reformat input affinities to 10^aff and diagonalize
+    # Sam: shouldn't this be done before the optimization?
     modelAffs = restructureAffs(monomerAffs)
 
     # Use the binding model to calculate bound receptors for target and off-target cell types
@@ -85,7 +86,6 @@ def minOffTargSelec(
 
 # Called in Figure1, Figure4, and Figure5
 def optimizeSelectivityAffs(
-    initialAffs: list,
     targRecs: np.ndarray,
     offTargRecs: np.ndarray,
     dose: float,
@@ -108,20 +108,29 @@ def optimizeSelectivityAffs(
     """
 
     # Relabel initial affinities to a variable that will be altered during optimization
-    X0 = initialAffs
+
 
-    # Set bounds for optimization
+    # Choose initial affinities and set bounds for optimization
     # minAffs and maxAffs chosen based on biologically realistic affinities for engineered ligands
     # Sam: affinities are maxing and bottoming out before optimization is complete...
-    #       for fig1, target 1, final affinities are 1e7 and ~1e9 (9.997e8)
+    #       for fig1, target 1, final affinities are 1e7 and ~1e9 (9.997e8) (with bounds 7 and 9)
     minAffs = [7.0] * (targRecs.shape[1])
     maxAffs = [9.0] * (targRecs.shape[1])
-    optBnds = Bounds(np.full_like(X0, minAffs), np.full_like(X0, maxAffs))
+    initAffs = np.full_like(
+        valencies[0], # Correct this if sizes of initial affinities and valencies are not always the same
+        minAffs[0] + (maxAffs[0] - minAffs[0]) / 2 # Start at midpoint between min and max bounds
+    )
+    optBnds = Bounds(
+        np.full_like(initAffs, minAffs), 
+        np.full_like(initAffs, maxAffs)
+    )
+    print(initAffs)
+    print(optBnds)
 
     # Run optimization to minimize off-target selectivity by changing affinities
     optimizer = minimize(
         minOffTargSelec,
-        X0,
+        initAffs,
         bounds=optBnds,
         args=(
             targRecs,
@@ -137,7 +146,6 @@ def optimizeSelectivityAffs(
     return optSelect, optAffs
 
 
-# Called in sampleReceptorAbundances
 # Sam: reimplement this function when we have a clearer idea 
 #   of how to calculate conversion factors
 def calcCITEConvFacts(CITE_DF: pd.DataFrame) -> pd.DataFrame: