Visualization and analysis of generated molecules

Once the generation is launched, the generated molecules can be visualized in real time by clicking on the eye icon in the generator box. If rewards that specified a TPP were selected, the user can also view the molecules matching the TPP (IN BLUEPRINT).

Parallel Coordinates

The molecules can be further filtered by dragging the mouse over the desired range of objective or metric in the Parallel Coordinates panel (see image below). Any filters applied to this panel can be saved using the Save button for future use. An example of Parallel Coordinates is shown in the below image:

The internal generation scores that are not present in the input data can be read as follows:

• QED

•  
  • A global drug-likeness metric, computed by RDKit https://www.rdkit.org/docs/source/rdkit.Chem.QED.html.

•  
  • Selection threshold: This score is largely informative, the values will strongly depend on the chemical space.

• Similarity

•  
  • Maximum Tanimoto similarity between the generated molecule and all the molecules of the chemical space. It is based on the Morgan fingerprints with 4096 bits and a radius of 4.

•  
  • Selection threshold: We recommend choosing Similarity > 0.6.

• Confidence

•  
  • In-house metric to make sure the predictors are within their applicability domain. It is based on the presence and absence of known structural features in the full initial training set. For instance, if no alcohol is present in the training set of the QSAR model, generating such functional group during a generation would trigger a low confidence. 

•  
  • Selection threshold: We recommend Confidence > 0.7.

• Quality

•  
  • In-house metric for molecular complexity. It helps avoid "ugly" generated compounds. It makes sure the generator does not deviate too much from drug-likeness in terms of important molecular descriptors (for example, the number of rotatable bonds). 

•  
  • Selection threshold: We recommend Quality > 0.7.

• Scaffold

•  
  • Whether or not the generic Murcko scaffold of the compound already exists in the generator chemical space.

• Iktos ranking
   

•  
  • Geometric mean of several scores (first transformed): Similarity, confidence, predictions of the selected QSAR models, 3D - structure or ligand based scores -, and API scores that are optimized during generation (note: Scores are transformed with a Gaussian transform before their geometric mean is computed). Post processed scores are not included in the Iktos ranking score. Gives a general idea of how well the molecule scores on several objectives.

• Goal Score

•  
  • The number of predicted objectives (QSAR scores, 3D - structure or ligand based scores - and API objectives) matched by the molecule, out of a scale of 10 (i.e. 10 * N_actives / N_scores). For example, if a generated molecule matches one out of two objectives then its goal score will be 5. If it matches both objectives, the score will be 10 and if it matches none then the score will be 0.

• Retrosynthesis score 

•  
  • Score calculated using our in-house software Spaya reflecting synthetic accessibility. The score is computed on a true retro-synthesis in post processing. It takes a bit of time (in the order of magnitude of a few seconds per molecule), so some time is necessary for all retrosynthesis scores to be retrieved. Generators that are stopped before their full completion sometimes will not have all retrosynthesis scores, with some molecules being scored as "NA". The higher the retrosynthesis score, the better. 

• Retrosynthesis steps

•  

  • Number of synthetic steps as calculated using our in-house software Spaya as another measure of synthetic accessibility. The lower the number of steps, the better.

Search Bar

Under the Parallel Coordinates panel is a search bar that allows the user to filter generated molecules by Makya ID or by substructures (using SMARTS, or the Sketcher tool accessible by clicking on the Pen icon).

NOTE: substructure search can take time depending on the number of results.

Example of a substructure search to keep molecules containing a cyclopropane ring:

Example of a substructure search by Makya ID to keep molecules #1 to #5000:

 

Generated Molecules

A grid of generated molecules can be seen under the Search bar. If any filters are applied, this grid is updated in real time. 
 

• See details:

The details about each molecule (closest molecule in chemical space, predicted objective values, retrosynthesis score and number of steps...) can be seen by clicking on the See Details button on the cell.

If the generator uses 3D parameters, you have access to 3D visualization windows in the "Docking" (for structure-based) or "3D shape" (for ligand-based) tabs.

If the generator is a fragment-based generator, a generation tree is also shown. The tree is not a retrosynthesis tree, but rather a representation of the number of modification steps (single-reactant or double-reactant) taken by the generator; to get a retrosynthesis route, we recommend pushing the resulting structure to Spaya, our retrosynthesis software. See the description of "number of generation steps" in the Fragment Growing presentation.

• Sorting:

Molecules can be sorted by any desired objective by clicking on Sort and choosing the sorting parameter.

• Radar plot:

A radar plot for all or a subset of metrics and objectives can be generated for each molecule, in its respective cell, by selecting the Radar Plots mode in the "View" module on the top right corner of the grid.

• Molecule Selection:

All molecules can be selected by clicking Select All in the "Selection" module at the top-right corner of the grid. Alternatively, individual molecules may be selected by clicking their box.

• Exporting Molecules:

Selected molecules can either be added to the cart, directly exported as a csv or sdf file (thus retaining 3D structure information when relevant), or can be used to create a new dataset (to be used for subsequent generations in Makya) by clicking Export in the top-right "Selection" module.

• Rescoring Molecules:

Selected molecules can be rescored using any Makya Scorer by clicking Rescore Selection in the top-right "Selection" module. The rescored molecules will be saved as a new Makya Dataset.
 

• Cart and Trash:

Molecules of interest can be put in the cart (accessible to all the project collaborators) by selecting them, and clicking on "Selection" / Add Selection to Cart. Molecules that an user don't want to see in the results can be put in the trash in the same manner. To review the cart or trash, select the corresponding option in the top-right "View" module.
 

• Copy:

Each molecule cell in the grid has a copy button to conveniently copy SMILES of the molecule.

• Spaya Button   

If Spaya (Iktos's AI-powered retrosynthesis tool) licence is included in the Makya deployment, each molecule has a Spaya button. Clicking this button redirects to the Spaya website (spaya.ai) and automatically begins a retrosynthetic route search for the molecule in the cell. To learn more about Spaya, please contact [e-mail]