NAMI is Now Live on Vecura: Simplify Large-Scale Chemical Library Clustering and Chemical Space Visualization

What is NAMI?

NAMI (N-Ary Mapping Interface) is an efficient cheminformatics tool designed for grouping vast collections of small molecules into structurally coherent clusters and projecting them into a low-dimensional chemical-space map. Developed as a sister project to MDANCE from the Miranda-Quintana lab, NAMI is optimized specifically for small-molecule datasets. It leverages the O(N) BitBIRCH hierarchical clustering algorithm on binary Morgan fingerprints to handle large-scale datasets with high speed and low memory usage.

It helps users easily navigate and compress vast chemical libraries into organized, distinct families of molecules. It is especially useful for virtual screening triage, finding structural series, eliminating redundancy in screening decks, and selecting diverse subsets of compound libraries.

What can users do with NAMI on Vecura?

With NAMI on Vecura, users can:

• Perform High-Throughput Molecular Clustering: Group thousands or millions of SMILES strings by structural similarity using Morgan circular fingerprints and the ultra-fast BitBIRCH algorithm.

• Generate 2D Chemical Space Coordinates: Compute cluster centroids and perform Principal Component Analysis (PCA) to project high-dimensional molecular fingerprints into two-dimensional (PC1, PC2) space.

• Customize Clustering Granularity: Adjust the similarity threshold (bb_threshold) and branching factors to control how tightly or broadly molecules are merged.

• Filter and Refine Visualizations: Filter out singletons or specify minimum/maximum cluster sizes to generate clear, noise-free chemical space overviews suitable for downstream plotting and analysis.

What the output means

The output provides clear structured files and dataset summaries, including a CSV of per-molecule cluster assignments (SMILES, Name, and cluster), a statistical summary of cluster size distribution, and PCA-mapped coordinates for each cluster centroid along with its closest representative SMILES.

This output should be used to support scientific decision making, such as selecting structurally diverse representatives for assays or prioritizing structural series. It does not replace experimental validation.

Why this matters

In modern drug discovery, screening libraries often consist of hundreds of thousands to millions of compounds. Navigating this vast chemical space to extract meaningful structural series or to design diverse screening decks presents a significant computational bottleneck. Traditional hierarchical clustering methods struggle to scale efficiently with large datasets due to quadratic time and space complexities.

NAMI addresses this bottleneck by using the BitBIRCH algorithm, which builds a highly compact Clustering Feature tree (CF-tree) in linear O(N) time. By bringing NAMI into Vecura's headless, cloud-hosted environment, researchers can execute these heavy computational pipelines seamlessly. They can instantly translate raw SMILES datasets into clean, visualizable structural maps without having to manage custom Python environments, local dependencies, or compute resources.

• Developed by: Miranda-Quintana Lab (mqcomplab)

• Source: NAMI GitHub Repository

• Reference: Official GitHub repository