Chemical Space Analysis

Dimensionality Reduction

Chemical space analysis begins with dimensionality reduction to visualize high-dimensional molecular data in 2D or 3D space, making it easier to identify patterns, clusters, and relationships between molecules.

Importing your molecules

Import molecules as SMILES by hovering on Import SMILES and choosing your preferred upload method. If you want to load a dataset that you have uploaded previously, press Load File instead.

Selecting features

Choose if you want Chemetrian to compute features for you (fingerprints, RDKit descriptors, etc), or choose "Use features from file" if your csv already contained features associated with your SMILES strings.

Choosing dimensionality reduction model

Decide on a dimensionality reduction model and distance metric. Hovering over section headers as well as choices will give additional information on them.

Choosing model hyperparameters

Select hyperparameter values if the model requires them.

Acquiring the projection

Press the Run Projection button. This can be as fast as a few seconds but will scale with the number of molecules uploaded.

Navigating the projection

Hovering will show molecular structures, and scrolling will zoom in or out of the map for more granular analysis of chemical space. Clicking a molecule's datapoint will keep the molecule selected. Click it again to deselect it. The molecule's SMILES string can be copied to the clipboard by pressing the copy button next to the SMILES string.

Customizing the projection

Clicking the gear symbol in the top right corner

Highlighting specific molecules

To highlight specific molecules on the structure, SMILES strings or lists of SMILES strings can be copied into the "search SMILES" box. Pressing enter will highlight them. Pressing over "Display searched SMILES" will give a list of SMILES that are highlighted. They can be unhighlighted here. Molecule highlight color can be changed by pressing the colored bar and selecting other colors.

Clusterization

Change the mode from "Projection" to "Clusterization".

Select clustering algorithm

Select a clustering algorithm and number of clusters if the algorithm requires it. Press "Run Clusterization" located underneath the projection.

The clustered projection

The clustered projection will appear. Each color represents a different cluster.

Silhouette scores

Silhouette score is a commonly used metric to evaluate cluster quality. Press "View Silhouette Score" to see a plot of scores across different numbers of clusters.

Silhouette plot

A plot like the one below will appear.

Molecules in each cluster

To obtain a list of which molecules belong to which clusters, press "Download Clusters".

Cluster Medoids

The medoid molecule of a cluster is the one most close to the middle. This can be interpreted as the medoid molecule being representative of a specific cluster within the dataset. Press "Download Medoids" for a list of medoid SMILES strings. Press "View Structures" to see the chemical structures of the Medoids.

Medoid structures

View the chemical structures of the representative medoid molecules for each cluster.

Navigation and customization

All the same navigation, customization, and highlighting tools as the dimensionality reduction are available for the clusterization plot.