Traditional Prediction Case

The results of the boiling point estimation using the neural network are shown below.

• The number of molecules used for the calculation: 1718
• The figure BP-1 shows the relationships of observed values and estimated values using the Joback method and our neural network method. The X-axis shows observed values, and the Y-axis shows estimated values.
  
  figure BP-1
  The approximated lines are
  Joback method : y = 1.0162x - 1.8794 R2 = 0.8945
  Neural Network : y = 1.0006x - 0.3138 R2 = 0.982
  
• The figure BP-2 shows the results of alkane's boiling point estimation. The X-axis shows carbon count, and the Y-axis shows boiling point values.
  
  figure BP-2

Some estimated examples for isomers (cis, trans, ortho, meta and para) are shown in this table.

The results of the logP estimation using the neural network are shown below.

• The number of molecules used for the learning: 425
• The number of molecules used for the test: 202
• Figure LP-1 and LP-2 show the relationships of observed values and estimated values using our neural network method. The X-axis shows observed values, and the Y-axis shows estimated values
• The figure LP-1 shows the learning results of the neural network.
  
  figure LP-1
  The approximated line is
  y = 0.9885x + 0.1334 R2 = 0.9903
• The figure LP-2 shows the estimated results of logP for test molecules.
  
  figure LP-2
  The approximated line is
  y = 0.9594x + 0.1334 R2 = 0.9109
• The comparison to other methods.
  

Some estimated and observed logP examples for isomers are shown in this table. (ortho, meta and para isomers)

The results of the solubility estimation using the neural network are shown below. The values are represented in a log scale of weight percent.

• The number of molecules used for the learning: 273
• The number of molecules used for the test: 58
• Figure S-1 and S-2 show the relationships of observed values and estimated values using our neural network method. The X-axis shows observed values, and the Y-axis shows estimated values.
• The figure S-1 shows the learning results of the neural network.
  
  figure S-1
  The approximated line is
  y = 1.0011x + 0.0076 R2 = 0.9635
• The figure S-2 is the estimated results of solubility for test molecules.
  
  figure S-2
  The approximated line is
  y = 0.9504 - 0.1795 R2 = 0.9032

Some estimated and observed solubility examples for isomers are shown in this table. (ortho, meta and para isomers)

The results of the pKa estimation using the neural network are shown below.

• The number of molecules used for the learning: 243
• The number of molecules used for the test: 67
• Figure PK-1 and PK-2 show the relationships of observed values and estimated values using our neural network method. The X-axis shows observed values, and the Y-axis shows estimated values.
• The figure PK-1 shows the learning results of the neural network.
  
  figure PK-1
  The approximated line is
  y = 0.9927x + 0.1068 R2 = 0.9900
• The figure PK-2 shows the estimated results of pKa for test molecules.
  
  figure PK-2
  The approximated line is
  y = 1.0494x - 0.0024 R2 = 0.9592

Some estimated and observed pKa examples for isomers are shown in this table. (ortho, meta and para isomers)

The results estimated Tg in water by the neural network are shown below.

• The number of molecules used for the learning: 45
• The number of molecules used for the test: 30
• Figure TG-1 and TG-2 show the relationships of observed values and estimated values using our neural network method. The X-axis shows observed values, and the Y-axis shows estimated values.
• The figure TG-1 shows the learning results of the neural network.
  
  figure TG-1
  The approximated line is
  y = 1.0004x + 1.5987 R2 = 0.9941
• The figure TG-2 shows the estimated results of pKa for test molecules.
  
  figure TG-2
  The approximated line is
  y = 0.9962x - 3.9485 R2 = 0.9685