def plot_diff(x_list: list, y_list: list):
fontsize = 27
from neutromeratio.analysis import compute_kl_divergence, bootstrap_rmse_r
plt.figure(figsize=[8, 8], dpi=300)
plt.subplot(211)
sns.distplot(x_list,
kde=True,
rug=True,
bins=15,
label="$\Delta G_{solv}^{exp}$")
sns.distplot(y_list,
kde=True,
rug=True,
bins=15,
label="$\Delta G_{solv}^{calc}$")
plt.xlabel("$\Delta G_{solv}$ [kcal/mol]", fontsize=fontsize)
plt.ylabel("Probability", fontsize=fontsize)
plt.axvline(0, 0, 15, color="red")
kl = compute_kl_divergence(np.array(x_list), np.array(y_list))
rmse, mae, rho = bootstrap_rmse_r(np.array(x_list), np.array(y_list), 1000)
plt.text(-13.0, 0.15, f"MAE$ = {mae}$", fontsize=fontsize)
plt.text(-13.0, 0.13, f"RMSE$ = {rmse}$", fontsize=fontsize)
plt.text(-13.0, 0.11, f"KL$ = {kl:.2f}$", fontsize=fontsize)
ax = plt.gca()
plt.setp(ax.get_xticklabels(), fontsize=18)
plt.setp(ax.get_yticklabels(), fontsize=18)
plt.legend(fontsize=fontsize)
plt.ylabel("Probability")
plt.yticks([])
plt.tight_layout()
plt.subplot(212)
sns.distplot(
np.array(x_list) - np.array(y_list),
kde=True,
rug=True,
bins=15,
label="$\Delta G_{solv}^{exp} - \Delta G_{solv}^{calc}$",
)
plt.xlabel("$\Delta G_{solv}^{exp} - \Delta G_{solv}^{calc}$ [kcal/mol]",
fontsize=fontsize)
plt.ylabel("Probability", fontsize=fontsize)
plt.axvline(0, 0, 15, color="red")
kl = compute_kl_divergence(np.array(x_list), np.array(y_list))
rmse, mae, rho = bootstrap_rmse_r(np.array(x_list), np.array(y_list), 1000)
plt.legend(fontsize=fontsize)
plt.yticks([])
plt.ylabel("Probability")
ax = plt.gca()
plt.setp(ax.get_xticklabels(), fontsize=18)
plt.setp(ax.get_yticklabels(), fontsize=18)
plt.tight_layout()
plt.show()
def plot_single_dist(x_list, y_list, label):
fontsize = 27
plt.figure(figsize=[8, 8], dpi=300)
plt.ylabel("Probability", fontsize=fontsize)
from neutromeratio.analysis import compute_kl_divergence, bootstrap_rmse_r
sns.distplot(
np.array(x_list) - np.array(y_list),
kde=True,
rug=True,
bins=15,
label="Boltzmann weighting - Minimum",
)
rmse, mae, rho = bootstrap_rmse_r(np.array(x_list), np.array(y_list), 1000)
kl = compute_kl_divergence(np.array(x_list), np.array(y_list))
# plt.text(-1.5, 1.5, f"RMSE$ = {rmse}$", fontsize=fontsize)
# plt.text(-1.5, 1.35, f"MAE$ = {mae}$", fontsize=fontsize)
# plt.text(8.0, 0.04, f"KL$ = {kl:.2f}$", fontsize=fontsize)
plt.xlabel(label, fontsize=fontsize)
plt.axvline(0, 0, 15, color="red")
# plt.legend(fontsize=fontsize)
plt.yticks([])
plt.show()
def plot_dist():
fontsize = 17
from neutromeratio.analysis import compute_kl_divergence, bootstrap_rmse_r
import seaborn as sns
sns.distplot(x_list,
kde=True,
rug=True,
bins=15,
label="befor optimization")
sns.distplot(y_list,
kde=True,
rug=True,
bins=15,
label="befor optimization")
rmse, mae, rho = bootstrap_rmse_r(np.array(x_list), np.array(y_list), 1000)
kl = compute_kl_divergence(np.array(x_list), np.array(y_list))
plt.text(8.0, 0.10, f"MAE$ = {mae}$", fontsize=fontsize)
plt.text(8.0, 0.09, f"KL$ = {kl:.2f}$", fontsize=fontsize)
plt.xlabel("$\Delta_{r}G_{solv}$", fontsize=fontsize)
plt.ylabel("Probability", fontsize=fontsize)
plt.show()
def analyse_optimization(base_path: str):
import seaborn as sns
plt.style.use("seaborn-deep")
from scipy.stats import entropy
from neutromeratio.analysis import compute_kl_divergence, bootstrap_rmse_r
all_names = _get_names()
# load the model perofrmance before optimization on the training/validation set
results_before_retraining = pickle.load(
open(base_path + "/results_before_training.pickle", "rb+"))
# results_before_retraining is a dictionary of lists, the keys are the mol names, the values are a list of [ddG, validation/training]
# load the ddG values for the test set before and after training
results_test_set = pickle.load(
open(base_path + "/results_AFTER_training_for_test_set.pickle", "rb+"))
# results_test_set is a dictionary of lists, the key are the mol anmes, the values are [experimental_ddG, before_optimization_ddG, after_optimization_ddG]
# get everythin in four lists
original_ = []
reweighted_ = []
exp_ = []
names_ = []
exp_results = _get_exp_results()
for n in results_test_set:
original, reweighted = results_test_set[n]
exp = (exp_results[n]["energy"] * unit.kilocalorie_per_mole
) # already in kcal/mol
names_.append(n)
original_.append(
(original * kT).value_in_unit(unit.kilocalorie_per_mole))
reweighted_.append(
(reweighted * kT).value_in_unit(unit.kilocalorie_per_mole))
exp_.append(exp)
# plot the distribution of ddG before and after optimization
sns.set(color_codes=True)
plt.figure(figsize=[8, 8], dpi=300)
fontsize = 25
delta_exp_original = np.array(exp_) - np.array(original_)
kl = compute_kl_divergence(np.array(exp_), np.array(original_))
rmse, mae, rho = bootstrap_rmse_r(np.array(exp_), np.array(original_),
1000)
plt.text(
-28.0,
0.175,
f"RMSE$ = {rmse}$",
fontsize=fontsize,
color=sns.xkcd_rgb["denim blue"],
)
plt.text(
-28.0,
0.16,
f"MAE$ = {mae}$",
fontsize=fontsize,
color=sns.xkcd_rgb["denim blue"],
)
plt.text(
-28.0,
0.145,
f"KL$ = {kl:.2f}$",
fontsize=fontsize,
color=sns.xkcd_rgb["denim blue"],
)
sns.distplot(
delta_exp_original,
kde=True,
rug=True,
bins=15,
label="ANI1ccx native",
color=sns.xkcd_rgb["denim blue"],
)
delta_exp_reweighted = np.array(exp_) - np.array(reweighted_)
rmse, mae, rho = bootstrap_rmse_r(np.array(exp_), np.array(reweighted_),
1000)
kl = compute_kl_divergence(np.array(exp_), np.array(reweighted_))
plt.text(
-28.0,
0.12,
f"RMSE$ = {rmse}$",
fontsize=fontsize,
color=sns.xkcd_rgb["pale red"],
)
plt.text(
-28.0,
0.105,
f"MAE$ = {mae}$",
fontsize=fontsize,
color=sns.xkcd_rgb["pale red"],
)
plt.text(
-28.0,
0.09,
f"KL$ = {kl:.2f}$",
fontsize=fontsize,
color=sns.xkcd_rgb["pale red"],
)
sns.distplot(
delta_exp_reweighted,
kde=True,
rug=True,
bins=15,
label="ANI1ccx optimized",
color=sns.xkcd_rgb["pale red"],
)
plt.legend(fontsize=fontsize - 5)
plt.ylabel("Probability", fontsize=fontsize)
plt.xlabel(
"$\Delta_{r} G_{solv}^{exp} - \Delta_{r} G_{vac}^{calc}$ [kcal/mol]",
fontsize=fontsize,
)
plt.xticks(fontsize=20)
plt.yticks(fontsize=20)
plt.yticks([])
plt.show()