Fix calibration after TL

Author: RobbinBouwmeester

CHANGELOG.md

@@ -6,6 +6,12 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to
 [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [3.1.1] - 2024-10-10
+
+### Changed
+
+- Revert to linear calibration after transfer learning only
+
 ## [3.1.0] - 2024-08-31
 
 ### Changed

deeplc/deeplc.py

@@ -610,7 +610,9 @@ def make_preds_core(
         try:
             X
             ret_preds = mod.predict(
-                [X, X_sum, X_global, X_hc], batch_size=self.batch_num_tf, verbose=int(self.verbose)
+                [X, X_sum, X_global, X_hc],
+                batch_size=self.batch_num_tf,
+                verbose=int(self.verbose),
             ).flatten()
         except UnboundLocalError:
             logger.debug("X is empty, skipping...")
@@ -832,27 +834,39 @@ def calibrate_preds_func_pygam(
         # measured_tr = list(measured_tr)
 
         # Fit a SplineTransformer model
-        spline = SplineTransformer(degree=4, n_knots=int(len(measured_tr) / 100) + 5)
-        spline_model = make_pipeline(spline, LinearRegression())
-        spline_model.fit(predicted_tr.reshape(-1, 1), measured_tr)
+        if self.deeplc_retrain:
+            spline = SplineTransformer(degree=2, n_knots=10)
+            linear_model = LinearRegression()
+            linear_model.fit(predicted_tr.reshape(-1, 1), measured_tr)
 
-        # Determine the top 10% of data on either end
-        n_top = int(len(predicted_tr) * 0.1)
+            linear_model_left = linear_model
+            spline_model = linear_model
+            linear_model_right = linear_model
+        else:
+            spline = SplineTransformer(
+                degree=4, n_knots=int(len(measured_tr) / 500) + 5
+            )
+            spline_model = make_pipeline(spline, LinearRegression())
+            spline_model.fit(predicted_tr.reshape(-1, 1), measured_tr)
 
-        # Fit a linear model on the bottom 10% (left-side extrapolation)
-        X_left = predicted_tr[:n_top]
-        y_left = measured_tr[:n_top]
-        linear_model_left = LinearRegression()
-        linear_model_left.fit(X_left.reshape(-1, 1), y_left)
+            # Determine the top 10% of data on either end
+            n_top = int(len(predicted_tr) * 0.1)
 
-        # Fit a linear model on the top 10% (right-side extrapolation)
-        X_right = predicted_tr[-n_top:]
-        y_right = measured_tr[-n_top:]
-        linear_model_right = LinearRegression()
-        linear_model_right.fit(X_right.reshape(-1, 1), y_right)
+            # Fit a linear model on the bottom 10% (left-side extrapolation)
+            X_left = predicted_tr[:n_top]
+            y_left = measured_tr[:n_top]
+            linear_model_left = LinearRegression()
+            linear_model_left.fit(X_left.reshape(-1, 1), y_left)
+
+            # Fit a linear model on the top 10% (right-side extrapolation)
+            X_right = predicted_tr[-n_top:]
+            y_right = measured_tr[-n_top:]
+            linear_model_right = LinearRegression()
+            linear_model_right.fit(X_right.reshape(-1, 1), y_right)
 
         calibrate_min = min(predicted_tr)
         calibrate_max = max(predicted_tr)
+
         return (
             calibrate_min,
             calibrate_max,

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "deeplc"
-version = "3.1.0"
+version = "3.1.1"
 description = "DeepLC: Retention time prediction for (modified) peptides using Deep Learning."
 readme = "README.md"
 license = { file = "LICENSE" }