code for figure creation (#3)

* FIGURES: port code for score distribution, add type checks

* FIGURES: admet satisfaction throughout AL iterations

* FIGURES: suppress RDKit warnings

* FIGURE: walltime figure

* FIGURES: add ability to change the metric for the indication of the upper boundary of the values for generated smiles (max or percentile)

* FIGURES: chemical similarity analysis for abl inhibitors

* STYLE: extract loader into a separate tool

* FIGURES: similarity analysis of scored distribution

* GIT: stop tracking exports folder

* test

* FIGURES: evolution of distribution analysis

* FIGURES: training curves

* FIGURES: pca analysis

* FIGURES: model memorization

* FIGURES: interactions count analysis

* FIGURES: dataset analysis

* FIGURES: dataset analysis

* FIGURES: all remaining figures code

.github/workflows/main.yml

@@ -2,7 +2,7 @@ name: Python application test with pytest
 
 on:
   push:
-    branches: [ main, sandbox ]
+    branches: [ main ]
 
 jobs:
   build:

.gitignore

@@ -6,6 +6,12 @@ sandbox.ipynb
 /PaperRuns/1_Pretraining/datasets/moses_train.csv.gz
 /PaperRuns/1_Pretraining/datasets/moses_test.csv.gz
 /PaperRuns/1_Pretraining/model_weights/
+/figures/exports/*/html/
+# /figures/exports/ligand_distribution/jpg/
+# /figures/exports/cluster_distribution/jpg/
+/figures/exports/
+secret.py
+*pkl
 
 # Byte-compiled / optimized / DLL files
 __pycache__/

ChemSpaceAL/Configuration.py

@@ -69,7 +69,7 @@
     "logP": {
         "func": lambda mol: Descriptors.MolLogP(mol),
         "lower": -0.4,
-        "upper": 6.5,
+        "upper": 6.5, # TODO: specify which Abl binder has log.p 6.3 
     },  # AdMET Lab recommends [0,3], [-0.4, 5.6] from Ghose
 }
 # Dictionary containing scores for different protein-ligand interactions

figures/admet_satisfaction.py

@@ -0,0 +1,85 @@
+import modules.secret
+import modules.Graph as Graph
+import modules.FilterPassing as flt_pass
+import pickle
+import os
+from typing import List
+
+GENERATIONS_PATH = modules.secret.PRODUCTION_RUNS_PATH + "2. Generation/smiles/"
+EXPORT_PATH = os.path.join(os.getcwd(), "figures", "exports", "admet_satisfaction", "")
+
+n_iters = 5
+ignored = {"fChar"}
+configs = [
+    ("model7_hnh_admet", "HNH", "ADMET", "softsub"),
+    ("model7_hnh_admetfg", "HNH", "ADMET+FGs", "softsub"),
+    ("model2_hnh", "HNH", "ADMET+FGs", "admetfg_softsub"),
+    ("model7_1iep_admet", "1IEP", "ADMET", "softsub"),
+    ("model7_1iep_admetfg", "1IEP", "ADMET+FGs", "softsub"),
+    ("model2_1iep", "1IEP", "ADMET+FGs", "admetfg_softsub"),
+]
+rerun_admet = False
+max_val = -float("inf")
+for prefix, target, filters, channel in configs:
+    print(prefix, target, filters, channel)
+    if rerun_admet:
+        fnames = flt_pass.prepare_generation_fnames(
+            prefix=prefix,
+            n_iters=n_iters,
+            channel=channel,
+            filters=filters,
+            target=target,
+        )
+        load_generation = flt_pass.prepare_generation_loader(base_path=GENERATIONS_PATH)
+        traces_lists: List[flt_pass.Trace] = []
+        filtered_dicts = []
+        max_val = -float("inf")
+        for i, fname in enumerate(fnames):
+            smiles = load_generation(fname)
+            filtToData = flt_pass.compute_admet_metrics(smiles)
+            filtered_dicts.append(filtToData)
+            pickle.dump(
+                filtered_dicts,
+                open(EXPORT_PATH + f"{prefix}_{filters}_{target}_dicts.pkl", "wb"),
+            )
+            traces, i_max_val = flt_pass.create_admet_metrics_traces(
+                filtToData,
+                showlegend=i == 0,
+                ignored_metrics=ignored,
+                distribution_upper_percentile=100,
+            )
+            max_val = max(max_val, i_max_val)
+            traces_lists.append(traces)
+    else:
+        filtered_dicts = pickle.load(
+            open(EXPORT_PATH + f"{prefix}_{filters}_{target}_dicts.pkl", "rb")
+        )
+        traces_lists = []
+        for i, filtToData in enumerate(filtered_dicts):
+            traces, i_max_val = flt_pass.create_admet_metrics_traces(
+                filtToData,
+                showlegend=i == 0,
+                ignored_metrics=ignored,
+                distribution_upper_percentile=95,
+            )
+            max_val = max(max_val, i_max_val)
+            traces_lists.append(traces)
+    max_val = 1.116
+    fig = flt_pass.create_admet_progression_figure(
+        traces_lists, v_space=0.1, h_space=0.08, y_max=max_val + 0.05
+    )
+    graph = Graph.Graph()
+    graph.update_parameters(dict(width=1000, height=700, annotation_size=28))
+    graph.style_figure(fig)
+    fig.update_layout(
+        showlegend=True,
+        legend=dict(
+            x=1.0,
+            y=0.5,
+            font=dict(size=18),
+        ),
+    )
+    graph.save_figure(
+        figure=fig, path=EXPORT_PATH, fname=f"{prefix}_{filters}_{target}"
+    )
+    # print(max_val)

figures/analyze_clustering.py

@@ -0,0 +1,33 @@
+import modules.secret
+import os
+import modules.ClusterAnalysis as ca
+
+SAMPLING_PATH = modules.secret.PRODUCTION_RUNS_PATH + "3. Sampling/kmeans_objects/"
+EXPORT_PATH = os.path.join(
+    os.getcwd(), "figures", "exports", "cluster_distribution", ""
+)
+
+import pickle
+
+
+configs = [
+    ("model7_1iep_admetfg", "1IEP", "ADMET+FGs", "softsub"),
+    # ("model2_1iep", "1IEP", "ADMET+FGs", "admetfg_softsub"),
+    # ("model7_1iep_admet", "1IEP", "ADMET", "softsub"),
+]
+n_iters = 5
+from modules.Graph import Graph
+
+for prefix, target, filters, channel in configs:
+    loader = ca.prepare_kmeans_fnames(filters="ADMET+FGs")
+    fnames = loader(prefix, n_iters, channel, filters, target)
+    fig = ca.plot_cluster_size_evolution(
+        path=SAMPLING_PATH, fnames=fnames, n_rows=2, n_cols=3
+    )
+    gr = Graph()
+    gr.save_figure(
+        figure=fig,
+        path=EXPORT_PATH,
+        fname=f"test_cluster_size_{prefix}_{channel}_{target}_{filters}",
+        html=True
+    )

figures/analyze_datasets_projection.py

@@ -0,0 +1,109 @@
+import modules.secret
+from modules.Graph import Graph
+import modules.DatasetAnalysis as dataset_analysis
+import modules.AnalyzeDistribution as analyze_dist
+import os
+from typing import cast, Iterable
+
+PRETRAINING_PATH = modules.secret.PRODUCTION_RUNS_PATH + "1. Pretraining/datasets/"
+GENERATIONS_PATH = modules.secret.PRODUCTION_RUNS_PATH + "2. Generation/smiles/"
+PCA_PATH = modules.secret.PRODUCTION_RUNS_PATH + "3. Sampling/pca_weights/"
+PCA_FNAME = "scaler_pca_combined_processed_freq1000_block133_120"
+EXPORT_PATH = os.path.join(os.getcwd(), "figures", "exports", "datasets_analysis", "")
+
+
+datasets = ["moses", "guacamol", "combined"]
+reduction = "PCA"
+desc_type = "mix"
+train_sample = 10_000
+generation_sample = 10_000
+
+training_smiles = [
+    dataset_analysis.load_training_smiles(
+        PRETRAINING_PATH, dataset, sample=train_sample
+    )
+    for dataset in datasets
+]
+generated_smiles = [
+    dataset_analysis.load_generated_smiles(
+        GENERATIONS_PATH, dataset, sample=generation_sample
+    )
+    for dataset in datasets
+]
+
+smile_set = set()
+for smile_container in [training_smiles, generated_smiles]:
+    for smiles in smile_container:
+        smile_set |= set(smiles)
+all_smiles = [training_smiles[i] + generated_smiles[i] for i in range(len(datasets))]
+# analyze_dist._calculate_descriptors_for_an_array(smiles=list(smile_set), save_path=EXPORT_PATH, save_name="moses_guac_combined", desc_mode=desc_type)
+descriptors = analyze_dist.load_descriptors(
+    load_path=EXPORT_PATH, load_fname="moses_guac_combined"
+)
+training_projection = [
+    analyze_dist.project_smiles(smiles, descriptors) for smiles in training_smiles
+]
+generated_projection = [
+    analyze_dist.project_smiles(smiles, descriptors) for smiles in generated_smiles
+]
+training_reduced, generated_reduced = dataset_analysis.reduce_training_and_generations(
+    training_projection, generated_projection, PCA_PATH, PCA_FNAME
+)
+
+# def invert_even_elts(array):
+#     return [array[1], array[0], array[3], array[2], array[5], array[4]]
+
+
+traces = dataset_analysis.prepare_scatter_traces(
+    training_reduced, generated_reduced,
+    labels=["Generations", "Training Set"],
+    colorscale=("#240046", "#80ed99"),
+    trace_opacity=0.5,
+    marker_size=2,
+    marker_width=0.1,
+)
+
+fig = dataset_analysis.plot_scatter2d_wsubplots(
+    traces=traces,
+    subplot_titles=[
+        "<b>MOSES</b>",
+        "<b>GuacaMol</b>",
+        "<b>Combined Dataset</b>",
+    ],
+    n_rows=1,
+    n_cols=3,
+)
+
+graph = Graph()
+graph.update_parameters(dict(
+        width=1100,
+        height=400,
+        xrange=[-18, 50],
+        yrange=[-18, 23],
+        xtick_len=4,
+        ytick_len=4,
+        xtick_width=1,
+        ytick_width=1,
+        axis_title_size=18,
+        xaxis_title="Principal Component 1 (18.6% variance explained)",
+        yaxis_title="Principal Component 2<br>(5.7% variance explained)",
+        show_xzero=True,
+        show_yzero=True,
+        annotation_size=20,
+    ))
+graph.style_figure(fig, force_annotations=False)
+fig.update_layout(legend=dict(
+        x=0.01,
+        y=1.0,
+        xanchor="left",
+        yanchor="top",
+        font=dict(size=16),
+        orientation="h",
+    ), yaxis2_title="", yaxis3_title="")
+graph.save_figure(
+    figure=fig,
+    path=EXPORT_PATH,
+    fname=f"{'+'.join(datasets)}_{desc_type}_{reduction}_trainsample{train_sample}_trainsample{generation_sample}",
+    html=False,
+    svg=True,
+)