By: Allyson Lynch
In [1]:
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import math
from scipy import stats
Plot the popularity distribution by count, cumulative probability, and log probability.
In [2]:
def popularity_plot(class_list, pareto_marker=False, log_plot=False, log_plot_reg=False, file=False):
"""takes a file with smiles strings and popularity calculation information for each category and the optional argument for a log plot and returns distribution plots"""
pareto_df = pd.read_table('Outputs/02 Popularity Files/pareto_split.tsv')
for molecule in class_list:
#make a dataframe per molcule (skip smiles)
smiles_df = pd.read_table('Outputs/02 Popularity Files/'+molecule+'_popularity.tsv')
smiles_df = smiles_df[smiles_df["number_" + molecule]>0] #remove if absent
graph_df = smiles_df.sort_values(by=["number_" + molecule]) #sort by count in increasing order
split_df = pareto_df[pareto_df['molecule']=='number_'+molecule]
#distribution plot
if file:
save_df = graph_df.loc[:, ['smiles', 'rank', "number_" + molecule]] #only include smiles, rank, and number
save_df.to_csv("Outputs/04 Graphs/popularity_plot/" + molecule + "distributionplot.tsv", sep='\t') #save as tsv
plt.figure() #seperate graphs
plot = sns.regplot(x= 'rank', y= "number_" + molecule, data= graph_df, fit_reg=False)
plot.set_title(molecule) #scatterplot of cumulative probabilities
fig = plot.get_figure()
fig.savefig("Outputs/04 Graphs/popularity_plot/"+molecule+"_distributionplot.svg")
#cumulative probability
if file:
save_df = graph_df.loc[:, ['smiles', molecule+"_proportion", molecule+"_rcumulativeprob"]] #only include smiles, rank, number, and cumulative count
save_df.to_csv("Outputs/04 Graphs/popularity_plot/" + molecule + "cumprobplot.tsv", sep='\t') #save as tsv
plt.figure() #seperate graphs
plot = sns.regplot(x= molecule+"_proportion", y= molecule+"_rcumulativeprob", data=graph_df, fit_reg=False)
if pareto_marker:
p=sns.regplot(x='percent_structures', y='percent_amines', data= split_df, fit_reg=False, marker="x", color='r')
p.text(x=split_df['percent_structures']+.03, y=split_df['percent_amines']-.03, s='Pareto split', fontsize=10, fontdict={'color':'r'})
plot.set_title(molecule) #scatterplot of amounts
fig = plot.get_figure()
fig.savefig("Outputs/04 Graphs/popularity_plot/"+molecule+"_cumprobplot.svg")
#log graph
if log_plot:
probability_df = graph_df.sort_values(by=[molecule + "_probability"], ascending=False) #sort by probability in decreasing order
x_values = [math.log(i, 10) for i in range(1, len(probability_df['smiles'].tolist())+1)] #list of rank
y_values = [math.log(i, 10) if i!=0 else 0 for i in probability_df[molecule + "_probability"].tolist()] #list of number
log_df = pd.DataFrame({'log rank':x_values, 'log probability':y_values}) #create dataframe
plt.figure()
if log_plot_reg:
plot = sns.regplot(x= 'log rank', y= 'log probability', data= log_df, fit_reg=True)
slope, intercept, r_value, p_value, std_err = stats.linregress(log_df['log rank'],log_df['log probability'])
print("y="+str(slope)+"x+"+str(intercept)+" r2="+str(r_value**2))
else:
plot = sns.regplot(x= 'log rank', y= 'log probability', data= log_df, fit_reg=False)
plot.set_title("log "+molecule) #scatterplot
fig = plot.get_figure()
fig.savefig("Outputs/04 Graphs/popularity_plot/"+molecule+"_logplot.svg")
Probability distribution of popularvsunpopular or successvsfailure for each property.
In [3]:
def pdf_graph(class_list, properties_file, outcome=False, include_absent_amines=True, file=False):
"""takes a popularity data file, properties data file, and optional parameter for including absent (-1) and returns a histogram of the probability distribution function"""
df_properties = pd.read_table(properties_file)
df_properties_drop = df_properties.drop(['_raw_Reaction number', '_raw_name', '_raw_SMILES', '_raw_SMILES.nosalt', '_rxn_Synthesis_Method', '_raw_Notes', '_out_Outcome'], axis=1)
if outcome:
df_popularity = pd.read_table("Outputs/03 Outcomes/triangle_outcomestats.tsv")
popularity_properties = pd.merge(df_popularity, df_properties, left_on='Smiles', right_on='Smiles', how='right') #merge popularity and properties by smiles
else:
df_popularity = pd.read_table("Outputs/02 Popularity Files/amine_popularity.tsv")
popularity_properties = pd.merge(df_popularity, df_properties, left_on='smiles', right_on='Smiles', how='right') #merge popularity and properties by smiles
popularity_properties[["metalborates_popular"]] = popularity_properties[["metalborates_popular"]].fillna(value=-1) #mark absent
#make a list per molecule...
for molecule in class_list:
#make popular and unpopular data frames
if outcome:
popular_df = popularity_properties.loc[popularity_properties['OUT_anySuccess?']==True] #success dataframe
unpopular_df = popularity_properties.loc[popularity_properties['OUT_anySuccess?']==False] #fail dataframe
else:
popular_df = popularity_properties.loc[popularity_properties[molecule+"_popular"]==1] #popular dataframe
if include_absent_amines:
unpopular_df = popularity_properties.loc[popularity_properties[molecule+"_popular"] <= 0] #unpopular + absent dataframe
else:
unpopular_df = popularity_properties.loc[popularity_properties[molecule+"_popular"]==0] #unpopular dataframe
#...of the pdfs for each property
for prop in df_properties.columns[1:-1]:
#make lists of property
discrete=False #intialize
popular = popular_df[prop].dropna().tolist()
unpopular = unpopular_df[prop].dropna().tolist()
if len(popular)>0:
if type(popular[0])==int:
discrete=True
if discrete:
if outcome:
graph_df = pd.DataFrame(columns=[prop, 'Probability', 'outcome'])
else:
graph_df = pd.DataFrame(columns=[prop, 'Probability', 'popularity'])
short = list(set(popular))
graph_df[prop] = short + short
if outcome:
graph_df['outcome'] = ['success' for i in range(len(short))] + ['failure' for i in range(len(short))]
else:
if include_absent_amines:
graph_df['popularity'] = ['popular' for i in range(len(short))] + ['notpopular' for i in range(len(short))]
else:
graph_df['popularity'] = ['popular' for i in range(len(short))] + ['unpopular' for i in range(len(short))]
graph_df['Probability'] = [popular.count(i)/len(popular) for i in short] + [unpopular.count(i)/len(unpopular) for i in short]
name = prop.replace("/", "_")
if file:
if outcome:
graph_df.to_csv("Outputs/04 Graphs/pdf/"+molecule+"_"+name+"_outcomepdf.tsv", sep='\t') #save as tsv
else:
graph_df.to_csv("Outputs/04 Graphs/pdf/"+molecule+"_"+name+"_pdf.tsv", sep='\t') #save as tsv
plt.figure() #seperate graphs
if outcome:
plot = sns.barplot(x=prop, y='Probability', hue='outcome', data=graph_df, palette=['g','r'])
else:
plot = sns.barplot(x=prop, y='Probability', hue='popularity', data=graph_df, palette=['k','0.50'])
plot.set(title=prop)
fig = plot.get_figure()
if outcome:
fig.savefig("Outputs/04 Graphs/pdf/"+molecule+"_"+name+"_outcomepdfplot.svg")
else:
fig.savefig("Outputs/04 Graphs/pdf/"+molecule+"_"+name+"_pdfplot.svg")
elif type(popular[0])==float:
plt.figure() #seperate graphs
if outcome:
plot = sns.distplot(popular, color='g')
plot = sns.distplot(unpopular, color='r')
else:
plot = sns.distplot(popular, color='k')
plot = sns.distplot(unpopular, color='0.75')
plot.set(title=prop)
fig = plot.get_figure()
prop = prop.replace("/", "_")
if outcome:
fig.savefig("Outputs/04 Graphs/pdf/"+molecule+"_"+prop+"_outcomepdfplot.svg")
else:
fig.savefig("Outputs/04 Graphs/pdf/"+molecule+"_"+prop+"_pdfplot.svg")
Plot the original and triangle manipulated property distribution.
In [4]:
def property_graph(properties_file, file=False):
"""takes a popularity data file, properties data file, and optional parameter for including absent (-1) and returns a histogram of the probability distribution function"""
df_properties = pd.read_table(properties_file, engine='python')
df_properties_drop = df_properties.loc[:, ['Smiles', '_rxn_pH', '_raw_Amine_Mol']]
#pH
graph_df = pd.DataFrame(columns=['_rxn_pH', 'Probability'])
values = df_properties['_rxn_pH'].tolist()
short = list(set(values))
graph_df['_rxn_pH'] = short
graph_df['Probability'] = [values.count(i)/len(values) for i in short]
if file:
graph_df.to_csv("Outputs/04 Graphs/property_graph/"+'_rxn_pH'+properties_file[20:-10]+"_barplot.tsv", sep='\t') #save as tsv
plt.figure() #seperate graphs
plot = sns.barplot(x='_rxn_pH', y='Probability', data=graph_df)
plot.set(title='_rxn_pH')
fig = plot.get_figure()
fig.savefig("Outputs/04 Graphs/property_graph/"+'_rxn_pH'+properties_file[20:-10]+"_barplot.svg")
#amine mols
values = df_properties['_raw_Amine_Mol'].tolist()
plt.figure()
plot = sns.distplot(values)
plot.set(title='_raw_Amine_Mol')
fig = plot.get_figure()
fig.savefig("Outputs/04 Graphs/property_graph/"+'_raw_Amine_Mol'+properties_file[20:-10]+"_barplot.svg")
In [5]:
def graphs(class_list, properties_files, pareto_marker=False, log_plot=False, log_plot_reg=False, include_absent_amines=True, outcome=False, file=False):
popularity_plot(class_list, pareto_marker, log_plot, log_plot_reg, file)
pdf_graph(class_list, properties_files[0], outcome, include_absent_amines, file)
for f in properties_files:
property_graph(f, file)
graphs(["metaloxides", "metalborates"], ["Outputs/03 Outcomes/triangle_clean.tsv", "Outputs/03 Outcomes/human_clean.tsv"], log_plot=True, outcome=True, file=True)
graphs(["metaloxides", "metalborates"], ["Outputs/03 Outcomes/triangle_clean.tsv", "Outputs/03 Outcomes/human_clean.tsv"], log_plot=True, file=True)
