Merge pull request #36 from choderalab/Add_Ki_Readout

Added Ki Readout

mtenn/config.py

@@ -60,6 +60,7 @@ class ReadoutConfig(StringEnum):
     """
 
     pic50 = "pic50"
+    ki = "ki"
 
 
 class CombinationConfig(StringEnum):
@@ -203,6 +204,8 @@ def build(self) -> mtenn.model.Model:
                 mtenn_pred_readout = mtenn.readout.PIC50Readout(
                     substrate=self.pred_substrate, Km=self.pred_km
                 )
+            case ReadoutConfig.ki:
+                mtenn_pred_readout = mtenn.readout.KiReadout()
             case None:
                 mtenn_pred_readout = None
 
@@ -211,6 +214,8 @@ def build(self) -> mtenn.model.Model:
                 mtenn_comb_readout = mtenn.readout.PIC50Readout(
                     substrate=self.comb_substrate, Km=self.comb_km
                 )
+            case ReadoutConfig.ki:
+                mtenn_comb_readout = mtenn.readout.KiReadout()
             case None:
                 mtenn_comb_readout = None
 

mtenn/readout.py

@@ -4,7 +4,8 @@
 
 
 class Readout(torch.nn.Module, abc.ABC):
-    pass
+    def __str__(self):
+        return repr(self)
 
 
 class PIC50Readout(Readout):
@@ -58,9 +59,6 @@ def __init__(self, substrate: Optional[float] = None, Km: Optional[float] = None
     def __repr__(self):
         return f"PIC50Readout(substrate={self.substrate}, Km={self.Km})"
 
-    def __str__(self):
-        return repr(self)
-
     def forward(self, delta_g):
         """
         Method to convert a predicted delta G value into a pIC50 value.
@@ -72,7 +70,7 @@ def forward(self, delta_g):
 
         Returns
         -------
-        float
+        torch.Tensor
             Calculated pIC50 value.
         """
         pic50 = -delta_g / torch.log(torch.tensor(10, dtype=delta_g.dtype))
@@ -81,3 +79,46 @@ def forward(self, delta_g):
             pic50 -= torch.log10(torch.tensor(self.cp_val, dtype=delta_g.dtype))
 
         return pic50
+
+
+
+class KiReadout(Readout):
+    """
+    Readout implementation to convert delta G values to Ki values. This new
+    implementation assumes implicit energy units, WHICH WILL INVALIDATE MODELS TRAINED
+    PRIOR TO v0.3.0.
+    Assuming implicit energy units:
+        deltaG = ln(Ki)
+        Ki = exp(deltaG)
+    """
+
+    def __init__(self):
+        """
+        Initialization.
+
+        Parameters
+        ----------
+        None
+        """
+        super(KiReadout, self).__init__()
+
+    def __repr__(self):
+        return f"KiReadout()"
+
+    def forward(self, delta_g):
+        """
+        Method to convert a predicted delta G value into a Ki value.
+
+        Parameters
+        ----------
+        delta_g : torch.Tensor
+            Input delta G value.
+
+        Returns
+        -------
+        torch.Tensor
+            Calculated Ki value.
+        """
+        ki = torch.exp(delta_g)
+
+        return ki