[pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

missense_kinase_toolkit/ml/src/esm2/analysis.py

@@ -1,38 +1,37 @@
 #!/usr/bin/env python
 
-import os
 import glob
+import os
+
+import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
-import matplotlib.pyplot as plt
 import seaborn as sns
+
 # os.chdir("/data1/tanseyw/projects/whitej/esm_km_atp/src")
-from utils import (
-    save_csv2csv, 
-    load_csv2dataset, 
-    parse_stats_dataframes,
-    invert_zscore,
-)
+from utils import invert_zscore, load_csv2dataset, parse_stats_dataframes, save_csv2csv
 
 path = "/data1/tanseyw/projects/whitej/esm_km_atp/"
 list_files = glob.glob(os.path.join(path, "*/fold-*.csv"))
 df = pd.read_csv(os.path.join(path, "assets/pkis2_km_atp.csv"))
 # log10 transform before z-scoring
-labels = df['ATP Conc.(uM)'].apply(np.log10)
+labels = df["ATP Conc.(uM)"].apply(np.log10)
 
 list_runs = [
     "5CV-KinCore-esm2_t6_8M_UR50D",
     "5CV-KLIFS_MIN-esm2_t6_8M_UR50D",
     "5CV-KLIFS_FULL-esm2_t6_8M_UR50D",
 ]
 
-list_logs = [glob.glob(os.path.join(path, run, "*/logs/fold-*.csv")) for run in list_runs]
+list_logs = [
+    glob.glob(os.path.join(path, run, "*/logs/fold-*.csv")) for run in list_runs
+]
 
 dict_logs = dict(zip(list_runs, list_logs))
 
 ### Load and process training and evaluation loss ###
 
-df_train, df_eval, df_final = pd.DataFrame(), pd.DataFrame(), pd.DataFrame()   
+df_train, df_eval, df_final = pd.DataFrame(), pd.DataFrame(), pd.DataFrame()
 for exp, list_file in dict_logs.items():
     for idx, file in enumerate(list_file):
         df_train_temp, df_eval_temp, df_final_temp = parse_stats_dataframes(file, idx)
@@ -45,7 +44,7 @@
 
 ### Load validation and training datasets ###
 
-#TODO: Look at all files
+# TODO: Look at all files
 ds_val, ds_train = load_csv2dataset(path, 5, "KLIFS_FULL_data.csv")
 df_val = pd.DataFrame()
 for idx, ds in enumerate(ds_val):
@@ -58,11 +57,13 @@
 list_fold = df_train["fold"].unique().tolist()
 list_replace = [f"Fold: {i}\n(n = {sum(df_val["fold"] != i)})" for i in list_fold]
 df_train["fold_label"] = df_train["fold"].map(dict(zip(list_fold, list_replace)))
-g = sns.FacetGrid(df_train, col="fold_label", row="exp", hue="exp", sharey=True, sharex=True)
+g = sns.FacetGrid(
+    df_train, col="fold_label", row="exp", hue="exp", sharey=True, sharex=True
+)
 g.map(sns.lineplot, "step", "loss")
 g.add_legend()
 g.set_axis_labels("Steps", "Training Loss")
-g.set_titles('{col_name}')
+g.set_titles("{col_name}")
 plt.savefig(os.path.join(path, "images/train_loss_2024.10.30.png"))
 
 # list_fold = df_train["fold"].unique().tolist()
@@ -80,7 +81,7 @@
 list_replace = [f"Fold: {i}\n(n = {sum(df_val["fold"] == i)})" for i in list_fold]
 df_eval["fold_label"] = df_eval["fold"].map(dict(zip(list_fold, list_replace)))
 df_eval["log_rmse"] = invert_zscore(df_eval["eval_rmse"], labels)
-df_eval["orig_rmse"] = df_eval["log_rmse"].apply(lambda x: 10 ** x)
+df_eval["orig_rmse"] = df_eval["log_rmse"].apply(lambda x: 10**x)
 
 # list_fold = df_eval["fold"].unique().tolist()
 # list_replace = [f"Fold: {i}\n(n = {sum(df_val["fold"] == i)})" for i in list_fold]
@@ -91,14 +92,16 @@
 # Leave RMSE in units of z-score log10(Km, ATP)
 
 sns.set(font_scale=1.5)
-df_eval["exp_label"] = df_eval["exp"].map(dict(zip(list_runs, ["KinCore KD", "KLIFS Pocket", "KLIFS Full Region"])))
+df_eval["exp_label"] = df_eval["exp"].map(
+    dict(zip(list_runs, ["KinCore KD", "KLIFS Pocket", "KLIFS Full Region"]))
+)
 g = sns.FacetGrid(df_eval, col="fold_label", hue="exp_label", sharey=True, sharex=True)
 # for ax in g.axes.flat:
 #     ax.axvline(500, color='r', linestyle='dashed', linewidth=1)
 g.grid(False)
 g.map(sns.lineplot, "step", "log_rmse")
 g.set_axis_labels("Steps", "Held-Out RMSE\n" + r"$(log_{10} K_{M, ATP})$")
-g.set_titles('{col_name}')
+g.set_titles("{col_name}")
 g.add_legend(title="Input sequence")
 # g.figsize(8, 6)
 plt.savefig(os.path.join(path, "images/eval_rmse_unconverted_2024.10.30.png"))
@@ -115,10 +118,10 @@
 
 g = sns.FacetGrid(df_eval, col="fold_label", hue="fold", sharey=False, sharex=False)
 for ax in g.axes.flat:
-    ax.axvline(500, color='r', linestyle='dashed', linewidth=1)
+    ax.axvline(500, color="r", linestyle="dashed", linewidth=1)
 g.map(sns.lineplot, "step", "orig_rmse")
 g.set_axis_labels("Steps", "RMSE, Eval (Converted)")
-g.set_titles('{col_name}')
+g.set_titles("{col_name}")
 plt.savefig(os.path.join(path, "images/eval_rmse_converted.png"))
 
 ### Plot histogram of labels for validation set ###
@@ -127,7 +130,7 @@
 list_replace = [f"Fold: {i}\n(n = {sum(df_val["fold"] == i)})" for i in list_fold]
 df_val["fold_label"] = df_val["fold"].map(dict(zip(list_fold, list_replace)))
 df_val["orig_label"] = invert_zscore(df_val["label"], labels)
-df_val["orig_label"] = df_val["orig_label"].apply(lambda x: 10 ** x)
+df_val["orig_label"] = df_val["orig_label"].apply(lambda x: 10**x)
 
 # Leave labels in units of z-score log10(Km, ATP)
 
@@ -137,9 +140,14 @@
 y, x, _ = plt.hist(df_val["label"])
 for idx, ax in enumerate(g.axes.flat):
     loc = df_val.loc[df_val["fold"] == idx + 1, "label"].mean()
-    ax.axvline(loc, color='r', linestyle='dashed', linewidth=1)
-    ax.text(loc + (x.max() - x.min()) * 0.1, y.max() * 0.9, "Mean: " + str(round(loc, 2)), color='r')
-g.set_titles('{col_name}')
+    ax.axvline(loc, color="r", linestyle="dashed", linewidth=1)
+    ax.text(
+        loc + (x.max() - x.min()) * 0.1,
+        y.max() * 0.9,
+        "Mean: " + str(round(loc, 2)),
+        color="r",
+    )
+g.set_titles("{col_name}")
 plt.savefig(os.path.join(path, "images/val_label_hist_zscore.png"), bbox_inches="tight")
 
 # Convert labels to original scale
@@ -150,16 +158,20 @@
 y, x, _ = plt.hist(df_val["orig_label"])
 for idx, ax in enumerate(g.axes.flat):
     loc = df_val.loc[df_val["fold"] == idx + 1, "label"].mean()
-    ax.axvline(loc, color='r', linestyle='dashed', linewidth=1)
-    ax.text(loc + (x.max() - x.min()) * 0.1, y.max() * 0.9, "Mean: " + str(round(loc, 2)), color='r')
-g.set_titles('{col_name}')
+    ax.axvline(loc, color="r", linestyle="dashed", linewidth=1)
+    ax.text(
+        loc + (x.max() - x.min()) * 0.1,
+        y.max() * 0.9,
+        "Mean: " + str(round(loc, 2)),
+        color="r",
+    )
+g.set_titles("{col_name}")
 plt.savefig(os.path.join(path, "images/val_label_hist_orig.png"), bbox_inches="tight")
 
 
-
 # import numpy as np
 # from utils import calc_zscore
 # df = pd.read_csv(os.path.join(path, "assets/pkis2_km_atp.csv"))
 # calc_zscore(df["ATP Conc.(uM)"].apply(np.log10))
 # df.head()
-# df["kd"].apply(len).max()
+# df["kd"].apply(len).max()

missense_kinase_toolkit/ml/src/esm2/batch_jobs.sh

@@ -12,4 +12,4 @@
 while IFS=, read -r model col_seq run_name
 do
     sbatch -J ${run_name} run.sh ${model} ${col_seq} ${run_name}
-done < batch_jobs.csv
+done < batch_jobs.csv

missense_kinase_toolkit/ml/src/esm2/inference.py

@@ -1,12 +1,9 @@
-import pandas as pd
-import seaborn as sns
 import matplotlib as mpl
 import matplotlib.pyplot as plt
+import pandas as pd
+import seaborn as sns
 import torch
-from transformers import (
-    AutoTokenizer, 
-    EsmForSequenceClassification
-)
+from transformers import AutoTokenizer, EsmForSequenceClassification
 
 path_model = "/data1/tanseyw/projects/whitej/esm_km_atp/5CV-KLIFS_MIN-esm2_t6_8M_UR50D/full/results/checkpoint-12500"
 
@@ -15,7 +12,9 @@
 model = EsmForSequenceClassification.from_pretrained(path_model).to(device)
 tokenizer = AutoTokenizer.from_pretrained("facebook/esm2_t6_8M_UR50D")
 
-df_klifs_zehir_muts_alphamissense = pd.read_csv("/data1/tanseyw/projects/whitej/esm_km_atp/assets/klifs_zehir_muts_alphamissense.csv")
+df_klifs_zehir_muts_alphamissense = pd.read_csv(
+    "/data1/tanseyw/projects/whitej/esm_km_atp/assets/klifs_zehir_muts_alphamissense.csv"
+)
 
 list_outputs = []
 for _, row in df_klifs_zehir_muts_alphamissense.iterrows():
@@ -25,26 +24,42 @@
 
 dict_outputs = dict(zip(df_klifs_zehir_muts_alphamissense["hgnc_name"], list_outputs))
 
-dict_muts = {i : None for i in dict_outputs.keys() if "_" in i}
+dict_muts = {i: None for i in dict_outputs.keys() if "_" in i}
 for key, value in dict_outputs.items():
     if "_" in key:
         wt = key.split("_")[0]
         dict_muts[key] = (dict_outputs[key] - dict_outputs[wt]) / dict_outputs[wt]
-
-df_klifs_zehir_muts_alphamissense["zscore_percent_change"] = df_klifs_zehir_muts_alphamissense["hgnc_name"].apply(lambda x: dict_muts[x] * 100 if x in dict_muts.keys() else None)
-df_klifs_zehir_muts_alphamissense["zscore_percent_change_log"] = df_klifs_zehir_muts_alphamissense["zscore_percent_change"].apply(lambda x: np.sign(x) * np.log10(np.abs(x)))
+
+df_klifs_zehir_muts_alphamissense["zscore_percent_change"] = (
+    df_klifs_zehir_muts_alphamissense["hgnc_name"].apply(
+        lambda x: dict_muts[x] * 100 if x in dict_muts.keys() else None
+    )
+)
+df_klifs_zehir_muts_alphamissense["zscore_percent_change_log"] = (
+    df_klifs_zehir_muts_alphamissense["zscore_percent_change"].apply(
+        lambda x: np.sign(x) * np.log10(np.abs(x))
+    )
+)
 
 
 sns.set(font_scale=2)
 sns.set_style(style="white")
 plt.figure(figsize=(20, 7))
 # ax = sns.scatterplot(data = df_klifs_zehir_muts_alphamissense, x = "alphamissense_score", y = "zscore_percent_change", hue = "alphamissense_class")
-ax = sns.scatterplot(data = df_klifs_zehir_muts_alphamissense, x = "alphamissense_score", y = "zscore_percent_change_log", hue = "alphamissense_class")
+ax = sns.scatterplot(
+    data=df_klifs_zehir_muts_alphamissense,
+    x="alphamissense_score",
+    y="zscore_percent_change_log",
+    hue="alphamissense_class",
+)
 # plt.axhline(y=0, color='red', linestyle='--')
 # plt.yscale('log')
-plt.legend(title = "Alphamissense Class")
+plt.legend(title="Alphamissense Class")
 plt.xlabel("Alphamissense Score")
 # plt.ylabel(" Predicted Z-score\n% Change vs. Wild-Type")
 plt.ylabel(r"$log_{10}$" + " Predicted Z-score\n% Change vs. Wild-Type")
-plt.savefig("/data1/tanseyw/projects/whitej/esm_km_atp/images/zscore_percent_change_vs_alphamissense_score_log.png", bbox_inches = "tight")
-# plt.savefig("/data1/tanseyw/projects/whitej/esm_km_atp/images/zscore_percent_change_vs_alphamissense_score.png", bbox_inches = "tight")
+plt.savefig(
+    "/data1/tanseyw/projects/whitej/esm_km_atp/images/zscore_percent_change_vs_alphamissense_score_log.png",
+    bbox_inches="tight",
+)
+# plt.savefig("/data1/tanseyw/projects/whitej/esm_km_atp/images/zscore_percent_change_vs_alphamissense_score.png", bbox_inches = "tight")