iChem Integration on Vecura: Instantaneous Chemical Library Similarity and Clustering

What is iChem?

iChem is a high-performance cheminformatics package developed by the Miranda-Quintana group at the University of Florida that implements instantaneous similarity (iSIM) to analyze compound sets. Unlike conventional methods that require computing expensive, slow pairwise similarity matrices scaling as $O(n^2)$, iChem generalizes classical similarity indices to operate on $n$ molecules simultaneously in a single, fast operation. It helps users analyze the overall diversity, group similarity, and internal structure of chemical libraries in milliseconds. It is especially useful for high-throughput screening preparation, identifying unique chemical scaffolds, and comparing massive molecular catalogs in early-stage drug-discovery pipelines.

What can users do with iChem on Vecura?

With iChem on Vecura, users can:

• Compute Group-level Similarity: Instantly calculate aggregate iSIM scores using Jaccard-Tanimoto, Russell-Rao, or Sokal-Michener indices to measure the internal chemical cohesion of a compound collection.
• Cluster Libraries Efficiently: Partition chemical libraries using the BitBIRCH hierarchical clustering algorithm, which works directly on packed binary fingerprints to automatically determine optimal clustering thresholds.
• Perform Cross-Library Comparisons: Execute direct cross-library comparisons using interiSIM or evaluate internal cohesion compared to a joint reference using intraiSIM to detect chemical space overlap.
• Find Medoids and Outliers: Identify the most representative (medoid) compound or the most unique (outlier) compound in any dataset using complementary-similarity scores to streamline candidate selection.

What the output means

The output provides quantitative similarity scores, cluster counts, and specific SMILES structures of representative compounds. Users receive an aggregate isim_value (ranging from 0 for complete diversity to 1 for identical sets), precise indices of clusters, and isolated SMILES for cluster medoids and library outliers.

This output should be used to support scientific decision making, such as selecting structurally diverse candidates for high-throughput assays or trimming down redundant screening libraries. It does not replace experimental validation.

Why this matters

Evaluating the diversity and overlap of compound libraries is a foundational task in chemical informatics and drug discovery. However, traditional pairwise similarity calculations suffer from quadratic $O(n^2)$ scaling, making the analysis of tens of thousands of compounds computationally prohibitive or requiring substantial local high-performance hardware. iChem's instantaneous similarity (iSIM) framework completely bypasses pairwise comparisons by working with column-wise bit sums of molecular fingerprint matrices. This results in sub-quadratic scaling that can evaluate large libraries in milliseconds.

By bringing iChem directly to Vecura, researchers can execute these state-of-the-art algorithms via an intuitive, guided workflow. This eliminates the need for managing Python environments, installing dependencies like RDKit, or maintaining local high-performance computing resources, ultimately accelerating the lead generation and compound optimization phases of research.

• Developed by: Miranda-Quintana group at the University of Florida
• Source: iChem GitHub Repository
• Reference: iSIM paper: Digital Discovery 2024