BTMS Science ROSHAMBO: Open-Source, Online Molecular Shape Comparison
2024
Biogen OS Chem ROSHAMBO GitHub
Shuwen Wang, Rasha Atwi, Ye Wang, Simone Sciabola, Adam Antoszewski1
Symposium 1
Biogen Medicinal Chemistry
Abstract Fig 2: Web App Visualizations Fig. 3: DUDE-Z Benchmarking
Efficient virtual screening techniques are critical in drug discovery for
identifying potential drug candidates. Molecular shape comparison
algorithms allow chemists to diversify existing chemical matter while
maintaining the shape of known-active molecules. We present
ROSHAMBO, an open-source package for molecular alignment and
3D similarity calculations optimized for large-scale virtual screening of
small molecules. Results demonstrated the package's near-state-of-
the-art performance and robustness across multiple target classes,
with speed that enables many routine ligand-based drug discovery
workflows. We have deployed ROSHAMBO for day-to-day use by
Biogen computational and medicinal chemists, and have made the
code freely available on GitHub. We have further developed an
interactive web app deployed on [Link], which allows
anyone to easily run ROSHAMBO and visualize results.
Computational Methods Features PAPER ROSHAMBO ROCS
CONCLUSIONS
Our open-source package employs RDKit1 to generate molecular Availability Open-Source Open-Source Commercial
conformations and the PAPER algorithm2 for optimizing molecular
alignments based on shape. After obtaining the optimal alignments GPU support ROSHAMBO is an open-source, publicly-available, and actively-
maintained molecular shape comparison algorithm that provides
between the target and the query molecules, both shape and color Shape/color scores - active compound enrichment at industrially-relevant speeds. It has a
(based on pharmacophore features) scores are computed to assess
molecular similarity. We output a list of drug-like molecules ranked by Conformer generation convenient Python API and command line interface to fit into most
their 3D similarity (Tanimoto or Tversky) to a known query. Analysis
computational workflows. It also includes tools that help not only to
visualize pharmacophores/overlaps, but also to benchmark and
Visualization BUY NOW BUY NOW BUY NOW
compare similar methods. The code, paired with thorough
Fig. 1: Algorithmic Workflow Command line interface documentation and an example Jupyter notebook, is available on
Query
Multiple overlap methods - GitHub and through the Biogen open-source web server (see QR
Screening
Database
Optional
Conformer Molecule
Shape Overlap
Optimization
codes above). This technology accelerates day-to-day medicinal
Generation
Initialization (Using a line search
algorithm)
chemistry workflows, including virtual screening and generative
Generation of
molecular design. Future generations of ROSHAMBO are in
Near state-of-the-art performance
N candidate
poses/database
…
molecules development, and promise a 1000x speedup, GPU parallelization,
Preprocessing
• Centering
The DUDE-Z dataset3 is comprised of ligands and decoys across a and increased accuracy, among other improvements.
ComboT ShapeT ColorT
Hits • Projecting
along PCA
Optimal diverse set of 43 industrially-relevant targets. These datasets are built
1.83 0.99 0.84 transformation
• Naming, …
arrays to have a 2% hit rate. When considering only the top 1% of molecules
1.36 0.85 0.51
scored by ROSHAMBO, we retrieve between 2 and 23 hits per REFERENCES
Similarity Calculations Molecule Transformations
Applying rotations and translations
dataset, for hit rates between 6% to 72%. ROSHAMBO thus provides
Aligned robust active molecule enrichment, and can process 73 (1) RDKit: Open-source cheminformatics. [Link]
structures
conformers/cpu/sec. Scaling to 24 CPUs and one NVIDIA GPU, a (2) Haque, I. S.; Pande, V.S. J Comp Chem 2010, 31 (1), 117-132.
user can screen over 6 million conformations per hour. (3) Stein, R. et al. J Chem Inf Mod 2021, 61, 699-714
