NEEMP:Reports

From WebChem Wiki
Jump to: navigation, search

Along with NEEMP we provide an handy python script to generate charge correlation graphs and quality assay reports, named nut-report.py.

SCRIPT BASIC USAGE:

Step 1:

  • Generation of chg-stats-out-file in which for each atom the difference between the reference QM charge and the EEM charge is calculated (figure 5). Such a file can be obtained running NEEMP in quality validation mode with the option --chg-stats-out-file (see example or quality validation for details).

Step 2:

  • Call nut-report.py with the above-mentioned file

./nut-report.py chg-stats-out-file

Requirements:

NB: the script has been tested for both Python 2.7 and Python 3.4, as well as for R 3.2.


Figure 5: Close-up from charge statistics file. Along with statistics for each molecule, ab-initio charges (3rd column), EEM charges (4th column) and their difference (5th column) are also printed out.
.

Once the script has been executed, a bunch of output files are generated in the chg-stats-out-file directory. In particular:

  • csv files containing per-atom charge information and values for several performance evaluating metrics for each atomic type and molecule
  • png files displaying the charge correlation graphs for the whole set and for each atomic type
  • html file gathering together all the previous information in an interactive and more easily readable report page

Figure 6 and Figure 7 compare few emblematic results extracted from the quality report files for two distinct NEEMP runs: the first evaluating a parameter set generated using the LR approach, meanwhile the second a parameter set generated using the DE-MIN approach. In both cases the training set is set02.sdf.

The following links provide access to the full reports:

Stats set02 de-summary.png
Stats set02 de-O1.png
Set02 de table.png
Figure 6: Left side: charge correlation graph for the whole set. Upper right side: example of charge correlation graph for a single atomic type (in this case oxygen presenting only single bonds). Lower right side: detail from the atomic types summary table, in which each row is coloured according to the RMSD column (red: high value, green: low value). In this particular case it is evident the high quality performance of the validated parameter set.
Stats set02 lr-summary.png
Stats set02 lr-O1.png
Set02 lr table.png
Figure 7: For a description of the figure layout refer to figure 6. It can be easily seen how the performance of the submitted parameter set is utterly poor. In specific the correlation graphs help to visualize the low degree of dependency between the EEM charges and the QM charges. Meanwhile the summary table provides the actual values of several statistical metrics and, as can be seen from the colouring pattern, the RMSD values are generally higher than the previous case.