--- title: "How to Cluster Peptides" slug: "how-to-cluster-peptides" tags: ["Discngine Ideation Analytics", "Peptides"] updated: 2026-09-24T08:48:55Z published: 2026-09-24T08:48:55Z canonical: "docs.discngine.com/how-to-cluster-peptides" --- > ## Documentation Index > Fetch the complete documentation index at: https://docs.discngine.com/llms.txt > Use this file to discover all available pages before exploring further. # How to Cluster Peptides Clustering groups the peptides in a report by structural similarity so you can see which families of analogs exist and how their properties differ. It is configured on the **Clusters** tab of the SAR Report settings drawer, and explored in the **Cluster Explorer** dialog — whose **Summary Table** is the main analysis view. Learn more For how the grouping is computed — what "similar" means, what the quality figures tell you, and why the scatter plot is not a picture of the clusters — see [Understanding Peptide Clustering](/public/docs/explanation-understanding-clustering). ## Step 1: Configure and run clustering Open **Settings → Clusters** and turn on **Enable clustering**. The **Run clustering** panel appears. ![The Clusters tab with clustering enabled, showing auto-detect optimal k, component weights, the Run Clustering button, and the resulting 8-cluster results chip](https://discngine.github.io/ideation/Images/how-to-cluster-peptides/step-1-configure-and-run-clustering-01.png) Set: - **Auto-detect optimal k** — let Ideation choose the number of clusters (by silhouette score), or turn it off and set **Number of clusters** with the slider. - **Component weights** — how much the **Backbone Monomers**, **Branches**, **Bridges**, and **Chemical Objects** each contribute to structural similarity. (Branches and Chemical Objects are disabled when the dataset has none.) Click **Run Clustering**. When it finishes, the **Clustering Results** section lists the clusters, each with its size and representative peptide (centroid). Click a cluster to filter the viewer to its members. ![The Clustering Results list showing 8 clusters with their sizes and centroid peptides, and the Explore Clusters button](https://discngine.github.io/ideation/Images/how-to-cluster-peptides/step-1-configure-and-run-clustering-02.png) ## Step 2: Open the Cluster Explorer Click **Explore Clusters** to open the full-screen **Cluster Explorer**. It has a **Clustering Configuration** panel on the left (adjust weights and re-run without leaving the dialog), tabbed views in the centre, and a **Cluster Selection / Filter** panel on the right that lists every cluster (size and medoid) and ends with **Apply To Viewer** and **Export All Clusters**. The tabs are: **Summary Table**, **UMAP - 2D View**, **UMAP - 3D View**, **Comparison Matrix**, **Property Distributions**, **Parallel Coordinates**, and **Silhouette Analysis**. ## Step 3: Read the Summary Table (the main view) The **Summary Table** tab (subtitle *"Cluster statistics comparison"*) is the most important view — it compares every cluster side by side, one cluster per row. ![The Cluster Explorer Summary Table showing the Cluster and Common pattern columns, with the cluster list and Apply To Viewer on the right](https://discngine.github.io/ideation/Images/how-to-cluster-peptides/step-3-read-the-summary-table-the-main-view-01.png) Read it column by column: | Column | What it tells you | | --- | --- | | **Cluster** | The cluster label with its colour (stays fixed while you scroll). | | **Common pattern** | The cluster's **consensus sequence** — the most common monomer at each position, plus shared bridges and chemical objects. A **Shared → Variable** color gradient marks how conserved each position is, and a `?` marks a variable position (< 60% conserved). This is the structural motif the cluster represents. | | **Quality** | The cluster's silhouette band — **Good** (≥ 0.5), **Fair** (≥ 0.25), or **Poor** — with the score. Higher means a tighter, better-separated cluster. | | **Peptides count** | How many peptides are in the cluster. | | **Outliers** | Members that fit poorly (negative silhouette) — a high count means a loosely defined cluster. | | **Monomers / Bridges / Chemical objects (avg)** | Average structural counts per cluster. | | **One box-plot per property** | The distribution of each numeric property, drawn on a shared scale down the column so clusters compare directly (with mean ± std). | Scroll the table right to reach the per-property box-plot columns: ![The Summary Table scrolled to the logP and pIC50 box-plot columns, one box-plot per cluster on a shared scale](https://discngine.github.io/ideation/Images/how-to-cluster-peptides/step-3-read-the-summary-table-the-main-view-02.png) Columns are sortable and resizable. ### Reading the common patterns 1. Scan the **Common pattern** column to identify the motif each cluster represents — this is the structural family the cluster captures. 2. Use **Quality** and **Outliers** to judge how tight each cluster is before trusting it. 3. Read a property's **box-plot column** top-to-bottom to link a structural pattern to an activity trend — for example a cluster whose `pIC50` box-plot sits higher than the others points to a promising sub-series. 4. Act on a finding: select the cluster and click **Apply To Viewer** to filter the canvas to it, or **Export All Clusters** to take the assignments away. ## Step 4: Use the other tabs The remaining tabs give complementary views. The **UMAP - 2D View** projects the peptides into two dimensions and colors them by cluster, giving a rough map of which peptides share composition (the **n_neighbors** and **min_dist** sliders tune the projection): ![The UMAP - 2D View tab showing the peptides projected into 2D and colored by cluster, with the n_neighbors and min_dist parameter sliders](https://discngine.github.io/ideation/Images/how-to-cluster-peptides/step-4-use-the-other-tabs.png) Do not judge cluster quality from this plot The projection is computed from a different, position-blind representation than the clustering itself, so a cluster may legitimately look scattered and two clusters may legitimately appear to overlap. Use the **Summary Table**'s quality column, which is derived from the real clustering distances. See [Understanding Peptide Clustering](/public/docs/explanation-understanding-clustering). **UMAP - 3D View** is the three-dimensional equivalent, **Comparison Matrix** shows inter-cluster distances, **Property Distributions** and **Parallel Coordinates** compare properties across clusters, and **Silhouette Analysis** shows per-cluster quality. Cluster selection inside the dialog stays local until you click **Apply To Viewer**. ## Next steps - [Understanding Peptide Clustering](/public/docs/explanation-understanding-clustering) — how the grouping and the quality figures are computed - [How to Analyze Hotspots and Synergies](/public/docs/how-to-analyze-hotspots-and-synergies) - [Understanding Peptides in Ideation](/public/docs/explanation-understanding-peptides)