Added zero-shot commands

niklases · Jan 4, 2024 · 61fa903 · 61fa903
1 parent 2c00079
commit 61fa903
Show file tree

Hide file tree

Showing 3 changed files with 105 additions and 8 deletions.
diff --git a/.vscode/launch.json b/.vscode/launch.json
@@ -16,8 +16,9 @@
                 "--help"
             ]
         },
+
         {
-            "name": "Python: PyPEF MKLSTS",
+            "name": "Python: PyPEF MKLSTS ANEH",
             "type": "python",
             "request": "launch",
             "env": {"PYTHONPATH": "${workspaceFolder}"},
@@ -30,7 +31,93 @@
                 "--wt", "${workspaceFolder}/datasets/ANEH/Sequence_WT_ANEH.fasta", 
                 "--input", "${workspaceFolder}/datasets/ANEH/37_ANEH_variants.csv"
             ]
-        }
+        },
+
+        {
+            "name": "Python: PyPEF MKLSTS avGFP",
+            "type": "python",
+            "request": "launch",
+            "env": {"PYTHONPATH": "${workspaceFolder}"},
+            "program": "${workspaceFolder}/pypef/main.py",
+            "console": "integratedTerminal",
+            "justMyCode": true,
+            "cwd": "${workspaceFolder}/datasets/AVGFP/",
+            "args": [
+                "mklsts", 
+                "--wt", "P42212_F64L.fasta", 
+                "--input", "avGFP.csv"
+            ]
+        },
+
+        {   // GREMLIN zero-shot steps:
+            // 1. $pypef param_inference --msa uref100_avgfp_jhmmer_119.a2m --opt_iter 100
+            // 2. $pypef hybrid -t TS.fasl --params GREMLIN
+            // or
+            // 2. $pypef hybrid -m GREMLIN -t TS.fasl --params GREMLIN
+            "name": "Python: PyPEF save GREMLIN avGFP model",
+            "type": "python",
+            "request": "launch",
+            "env": {"PYTHONPATH": "${workspaceFolder}"},
+            "program": "${workspaceFolder}/pypef/main.py",
+            "console": "integratedTerminal",
+            "justMyCode": true,
+            "cwd": "${workspaceFolder}/datasets/AVGFP/",
+            "args": [
+                "param_inference", 
+                "--msa", "uref100_avgfp_jhmmer_119.a2m", 
+                "--opt_iter", "100"
+            ]
+        },
+
+        {
+            "name": "Python: PyPEF hybrid/only-TS-zero-shot GREMLIN-DCA avGFP",
+            "type": "python",
+            "request": "launch",
+            "env": {"PYTHONPATH": "${workspaceFolder}"},
+            "program": "${workspaceFolder}/pypef/main.py",
+            "console": "integratedTerminal",
+            "justMyCode": true,
+            "cwd": "${workspaceFolder}/datasets/AVGFP/",
+            "args": [
+                "hybrid", 
+                //"-m", "GREMLIN",   // optional, not required  
+                "--ts", "TS.fasl", 
+                "--params", "GREMLIN"
+            ]
+        },
 
+        {   // PLMC zero-shot steps:
+            // 1. $pypef param_inference --params uref100_avgfp_jhmmer_119_plmc_42.6.params
+            // 2. $pypef hybrid -t TS.fasl --params PLMC
+            "name": "Python: PyPEF save PLMC avGFP model",
+            "type": "python",
+            "request": "launch",
+            "env": {"PYTHONPATH": "${workspaceFolder}"},
+            "program": "${workspaceFolder}/pypef/main.py",
+            "console": "integratedTerminal",
+            "justMyCode": true,
+            "cwd": "${workspaceFolder}/datasets/AVGFP/",
+            "args": [
+                "param_inference", 
+                "--params", "uref100_avgfp_jhmmer_119_plmc_42.6.params"
+            ]
+    },
+
+        {   
+            "name": "Python: PyPEF hybrid/only-TS-zero-shot PLMC-DCA avGFP",
+            "type": "python",
+            "request": "launch",
+            "env": {"PYTHONPATH": "${workspaceFolder}"},
+            "program": "${workspaceFolder}/pypef/main.py",
+            "console": "integratedTerminal",
+            "justMyCode": true,
+            "cwd": "${workspaceFolder}/datasets/AVGFP/",
+            "args": [
+                "hybrid", 
+                "--ts", "TS.fasl", 
+                "--params", "PLMC",
+                "--threads", "24"
+            ]
+        }
     ]
 }
diff --git a/README.md b/README.md
@@ -30,7 +30,7 @@ Preprint available at bioRxiv: https://doi.org/10.1101/2022.06.07.495081.
   - [Model Hyperparameter Grids for Training](#model-hyperparameter-grids-for-training)
   - [Setting Up the Scripts Yourself](#setting-up-the-scripts-yourself)
   - [Preprocessing for DCA-based Sequence Encoding](#preprocessing-for-dca-based-sequence-encoding)
-  - [Unsupervised (DCA-based) zero-shot prediction](#unsupervised-dca-based-zero-shot-prediction)
+  - [Unsupervised/zero-shot prediction](#unsupervisedzero-shot-prediction)
   - [API Usage for Sequence Encoding](#api-usage-for-sequence-encoding)
 ---
 
@@ -412,15 +412,22 @@ python3 ./pypef/main.py
     ```
 
 <a name="zero-shot-prediction"></a>
-## Unsupervised (DCA-based) zero-shot prediction
+## Unsupervised/zero-shot prediction
 Several developed methods allow unsupervised prediction of a proteins fitness based on its sequence (and/or structure).
 These methods have the advantage that no initial knowledge about a proteins fitness is required for prediction while a correlation of the predicted score and a protein's natural fitness is assumed.
-DCA itself was a statistical/unsupervised method based on MSA information that outperforms simpler MSA-based methods (such as (un)coupled raw MSA sequence frequencies or BLOSUM scores), see [scripts/GREMLIN_numba/using_gremlin_functionalities.ipynb](scripts/GREMLIN_numba/using_gremlin_functionalities.ipynb).
-To make zero-shot predictions using PyPEF (plmc-DCA or GREMLIN-DCA) just do not provide a train set for model testing and use the DCA encoding method, e.g.
+DCA itself is a statistical/unsupervised method based on MSA information that outperforms simpler MSA-based methods (such as (un)coupled raw MSA sequence frequencies or BLOSUM scores), e.g., see [scripts/GREMLIN_numba/using_gremlin_functionalities.ipynb](scripts/GREMLIN_numba/using_gremlin_functionalities.ipynb).
+To make zero-shot predictions using PyPEF (plmc-DCA or GREMLIN-DCA) just do not provide a train set for model testing and apply the DCA encoding method for running, e.g., for the avGFP data,
 
 ```
-TODO
+pypef param_inference --msa uref100_avgfp_jhmmer_119.a2m
+pypef hybrid -t AVGFP_TS.fasl --params GREMLIN
 ```
+using the GREMLIN parameters, or,
+```
+pypef param_inference --params uref100_avgfp_jhmmer_119_plmc_42.6.params
+pypef hybrid -t TS.fasl --params PLMC
+```
+using the plmc parameters.
 
 Other well-performing zero-shot prediction methods with available source code are (list not complete, see ProteinGym [repository](https://github.com/OATML-Markslab/ProteinGym) and [website](https://proteingym.org/) for a more detailed overview of available methods and achieved performances):
 - ESM-1v/ESM-2 (https://github.com/facebookresearch/esm)

diff --git a/pypef/dca/hybrid_model.py b/pypef/dca/hybrid_model.py
@@ -576,7 +576,10 @@ def get_model_path(model: str):
         elif isfile(f'Pickles/{model}'):
             model_path = f'Pickles/{model}'
         else:
-            raise SystemError("Did not find specified model file.")
+            raise SystemError("Did not find specified model file in current working directory "
+                              " or /Pickles subdirectory. Make sure to train/save a model first "
+                              "(e.g., for saving a GREMLIN model, type \"pypef param_inference --msa TARGET_MSA.a2m\" "
+                              "or, for saving a plmc model, type \"pypef param_inference --params TARGET_PLMC.params\").")
         return model_path
     except TypeError:
         raise SystemError("No provided model. "