GPU-Accelerated Virtual Screening: ROSHAMBO is Now Available on Vecura

What is ROSHAMBO?

ROSHAMBO is an open-source, GPU-accelerated Gaussian molecular shape comparison tool designed for rapid, ligand-based virtual screening. By coupling the high-performance PAPER GPU kernel with RDKit-based conformer generation, it performs massive, parallel evaluations of analytic or Gaussian volume integrals. This allows the tool to rigidly align large libraries of candidate molecules to a reference query 3D shape, offloading computationally expensive mathematical tasks to GPUs.

It helps users calculate ShapeTanimoto and ColorTanimoto similarity scores for millions of chemical compounds in a fraction of the time required by traditional CPU methods. It is especially useful for hit-enrichment campaigns in early-stage drug discovery where researchers need to efficiently identify promising, structurally diverse lead candidates from massive screening datasets.

What can users do with ROSHAMBO on Vecura?

With ROSHAMBO on Vecura, users can:

• Screen libraries against a reference shape: Upload a query reference molecule (SDF or SMILES) and run rigid-body 3D shape matching against millions of candidate compounds.

• Generate conformers on the fly: Create high-quality 3D conformations of dataset molecules using RDKit's ETKDG pipeline directly within the screening workflow.

• Incorporate pharmacophoric scoring: Combine traditional 3D shape matching with electrostatic and hydrogen-bonding similarity (ColorTanimoto) to compute an informative ComboTanimoto score.

• Optimize alignment properties: Tweak parameters like ignoring hydrogen atoms, choosing exact analytical vs. fast Gaussian volume-overlap integrals, or enforcing carbon-only atomic radii to fine-tune alignment speed and precision.

What the output means

The output provides a ranked hit table and fully aligned 3D molecular structures. The hits_table details ROCS-compatible similarity metrics—such as ShapeTanimoto, ColorTanimoto, ComboTanimoto (ranging from 0 to 2), Fit/Ref Tversky variants, and raw overlap volume—sorted descending by your chosen score column. Additionally, the aligned_hits_sdf file provides the candidate molecules in their optimized 3D-superposed coordinate states, ready for direct visual inspection in molecular viewers or ingestion into downstream molecular docking and QSAR modeling workflows.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

In computer-aided drug design, ligand-based virtual screening remains an essential strategy to mine massive compound libraries for novel therapeutics, particularly when high-resolution structural details of the target protein are unavailable. Evaluating 3D shape similarity provides a powerful, physics-based heuristic: molecules sharing similar three-dimensional volume envelopes and pharmacophoric feature layouts are highly likely to interact with the same biological binding pockets. However, executing millions of rigid-body alignments and evaluating three-dimensional Gaussian overlap integrals is a massive computational bottleneck, historically requiring expensive commercial software licenses and extensive CPU clusters.

ROSHAMBO breaks down these barriers by delivering an open-source, enterprise-grade alternative powered by the massively parallel PAPER GPU kernel. By bringing ROSHAMBO to Vecura, researchers gain instant access to these advanced GPU capabilities without the hassle of configuring local CUDA environments, compiling complex libraries, or managing hardware constraints. This unified, cloud-accelerated approach enables rapid, scalable virtual screening campaigns, drastically shortening the time-to-insight for identifying high-value drug leads.

• Developed by: Molecular Informatics Research Group

• Source: GitHub

• Reference: ROSHAMBO Paper — J. Chem. Inf. Model. 2024 | Official GitHub Repository