1. Double-click the Materials Science icon
   • (No icon? See Starting Maestro)

Figure 2-1. Change Working Directory option.

2. Go to File > Change Working Directory
3. Find your directory, and click Choose
4. Pre-generated input and results files are included for running jobs or examining output. Download the zip file here: schrodinger.com/sites/default/files/s3/release/current/Tutorials/zip/opto_al.zip
5. After downloading the zip file, unzip the contents in your Working Directory for ease of access throughout the tutorial

Figure 2-2. Save Project panel.

6. Go to File > Save Project As
7. Change the File name to optoelectronics_AL_tutorial, click Save
   • The project is now named optoelectronics_AL_tutorial.prj

Figure 2-3. Importing the iridium complexes.

For this tutorial, we will study a relatively small data set of octahedral iridium complexes for example purposes. In general, the active learning and prediction workflow is typically performed with larger data sets. Let’s proceed to import the structures:

8. Go to File > Import Structures
9. Navigate to where you downloaded the tutorial files (if not already in your Working Directory) and select iridium_complexes.mae
10. Click Open
    • A new entry group is added to the entry list containing 249 entries

Note: These structures have not been optimized at the QM level. The workflow utilized herein will include QM optimizations on a subset of the data.

Figure 3-1. Selecting the entry group and opening the panel.

1. Select(1) the atoms are chosen in the Workspace. These atoms are referred to as "the selection" or "the atom selection". Workspace operations are performed on the selected atoms. (2) The entry is chosen in the Entry List (and Project Table) and the row for the entry is highlighted. Project operations are performed on all selected entries the entire optoelectronic_al_Ir_complexes (249) entry group from the entry list
   • Recall that selection means to highlight all of the entries in the group, which can be accomplished by clicking on the entry group name
2. Go to Tasks > Materials > Informatics > Optoelectronics Active Learning Training
   • The Optoelectronics Active Learning Training panel opens

Figure 3-2. Selecting a property.

3. In the Property section, choose First triplet Energy and Targeted Value
   • This indicates that we wish to target complexes with triplet energies near a specific value
4. Click Add
   • Triplet Energy - Targeted Value appears in the Multi-property Optimization (MPO) section of the panel

Note: There are many properties available in the Optoelectronics Active Learning panel. For more on these properties, please visit the help documentation.

Figure 3-3. Defining the property to be optimized.

5. Set the Targeted value to 2.80 eV
6. Set the Inner tolerance to 0.20 eV, the Outer tolerance to 0.40 eV and retain the Weight of 1.00
   • These quantities indicate which triplet energies will be considered as ‘good’ and ‘bad’ by the model, which we will describe in more detail shortly

Figure 3-4. Defining the Training parameters.

In the Training parameters section:

7. Set the Initial set size to 10
8. Set Additional compounds per iteration to 5
9. For Stop training if, choose The number of iterations reaches and input 15

These settings are used to parameterize the active learning training protocol. QM calculations will be performed on ten compounds initially. For each subsequent loop, the top five compounds identified by the ML algorithm will be used as input for QM calculations. The looping procedure will stop after 15 iterations.

Figure 3-5. Preparing and running the job.

10. Change the Job name to optoelectronics_al_iridium

This job takes several days on a 10+ CPU host. It is not necessary to run the job for this tutorial. However, if you would like to, adjust the job settings () as needed and click Run

11. Otherwise, close the Optoelectronic Active Learning Training panel and we will proceed to import pre-generated results

Figure 4-1. Importing the pregenerated results.

1. From the main menu, go to File > Import Structures
2. Navigate to the provided files and choose the Section_04 > optoelectronics_al_iridium > optoelectronics_al_iridium-out.maegz file
3. Click Open
   • A new entry group is added to the entry list entitled optoelectronics_al_iridium-out1 (249) containing the same 249 entries as the original data set

Figure 4-2. Analyzing the Project Table.

Various properties from the job are now associated with the output entries and can be visualized in the Project Tabledisplays the contents of a project and is also an interface for performing operations on selected entries, viewing properties, and organizing structures and data

4. Open the Project Table ()

Two new properties are displayed by default: optoelal MPO Score and optelec Triplet Energy (eV)

The former refers to the MPO score for each entry predicted from the final machine learning model from the active learning protocol. The latter is the triplet energy calculated by DFT for any entries that underwent QM calculations.

Figure 4-3. Adding and viewing additional properties from the Property Tree.

5. Go to the Property Tree (), expand All > Materials Science > Secondary and check optoelal iteration and optoelal MPO Score DFT
   • The two additional properties are added to the Project Table
   • optoelal iteration refers to the iteration in the active learning protocol in which the entry was selected for DFT calculations (if any)
   • MPO Score DFT refers to the MPO score calculated for that entry using the triplet energy from the QM job rather than the machine learning algorithm

Note: You can export directly from the Project Table to spreadsheet form if needed

6. Close the Project Table before proceeding

Figure 4-4. Importing the ML models generated by the active learning loop.

This active learning protocol can be used to both a) find the ‘best’ molecules from a data set and b) efficiently develop ML models. To visualize the models and make predictions, the Optoelectronic Active Learning Prediction panel is used

7. Go to Tasks > Materials > Informatics > Optoelectronics Active Learning Prediction
   • The Review and Apply Optoelectronics Active Learning Model panel opens
8. Click Load models

Figure 4-5. Selecting model files.

9. Navigate to the provided files and choose all of the Section_04 > optoelectronics_al_iridium > optoelectronics_al_iridium_mpo_*.alomgz files, where the * represents each iteration of the active learning loop
10. Click Open

Figure 4-6. Viewing the MPO ML models.

All fifteen ML models (one per iteration) are loaded into the panel. For each model, you can see the various parameters of interest, as well as view the corresponding scatter plot.

It is important to note that the ‘best’ models can sometimes be located before the final iteration. In this example, we will choose a model with a high R² for both the test and training sets as our ‘best’ model. In practice, please choose the model based on your research needs. In this case, it appears that the sixth iteration (optoelectronics_al_iridium_mpo_6.alomgz) gives the best ML model for predicting MPO score (R² (training) = 0.904, R² (test) = 0.895; these values are discussed in the help documentation).

It is also important to note that these models are for predicting MPO score as opposed to triplet energy directly.

Figure 4-7. Scatter plot for an MPO ML model.

11. Click Show for optoelectronics_al_iridium_mpo_6.alomgz
    • The scatter plot is shown. The predicted data are the MPO scores based on the ML algorithm and the trained data are the MPO scores as determined by the DFT calculated properties for all of the compounds preceding this iteration
12. Click OK to close the scatter plot

Figure 4-8. Loading a triplet energy ML model.

We can also view the ML models for predicting triplet energy directly. For example:

13. Click Load models
14. Navigate to the provided files and choose the optoelectronics_al_iridium_triplet_6.alomgz file
15. Click Open
    • An additional model is added to the table

Figure 4-9. Scatterplot for a triplet energy ML model.

16. Click Show for optoelectronics_al_iridium_triplet_6.alomgz

This ML model is for directly predicting triplet energy, and the corresponding scatter plot can again be visualized, this time comparing predicted to calculated triplet energy.

17. Click OK to close the scatter plot

Figure 4-10. Plotting the Learning Curves.

18. Select all 15 MPO models (Shift + Click)
19. Click Plot Learning Curves

Figure 4-11. Viewing the Learning Curves.

The Learning Curves plot displays the R² (Training) and R² (Test) for the ML model at each iteration of the active learning loop. This learning curve is useful for checking the convergence of a model and choosing which is the ‘best’. In this Figure we plot the MPO learning curves, but we could also plot the curves directly for our property of interest. Here we see that after iteration 2, the best MPO does not change. The top scoring compound remains the same for the rest of the iterations.

20. Click OK

Figure 4-12. A top candidate structure based on the active learning protocol.

Proceed to explore any of the models as you wish. At this point, there are a few possible next steps. You could proceed to:

• Scan the Project Table for the compounds that best match your target triplet energy.

Optional: To facilitate searching, you can sort the project table by clicking the arrow under the “optelec Triplet Energy (eV)” column and clicking “Sort All (Ascending)”

Figure 4-13. Using the panel for prediction.

• Optional: Use one of the top models to make predictions on a new set of compounds. To do so, simply select the structures from the entry list, choose a model from the Optoelectronic Active Learning Prediction panel and run the job