RIDE: A Next-Generation 3D Ligand-Based Screening Method

RIDE (Rapid Isostere Discovery Engine) is a powerful ligand-based virtual screening tool developed by MolSoft. It is designed to perform ultra-fast 3D isostere screening using MolSoft’s proprietary APF (Atomic Property Fields) technology. RIDE enables rapid identification of novel chemotypes (scaffold hopping) with similar 3D pharmacophoric and physicochemical features. RIDE allows you to screen 1.5 million conformers/sec/GPU, if you have a CPU you can screen 6000 conformers/sec/CPU. 

RIDE employs a systematic search algorithm with a high heuristic focus on searching the relevant space. RIDE uses pre-generated conformers and field screening based on translation and rotation. The RIDE (CPU) version is faster on multicore, and RIDE (GPU) is a lightning-fast, screening ~ 1.5 million conformers/second/GPU (RTX-4090).

RIDE can be run inside the ICM-Pro GUI or via a simple web-based interface, providing ease of use for chemists. Additionally, a batch script may also be used. In either format, RIDE allows fine-tuning of queries to generate the most desirable hit lists. This includes:

• Atom Weighting – Contributions of different portions of the molecule can be modulated with per-atom weights to reflect the relative importance of certain moieties.
• Excluded Volumes meet Shape Matching – An envelope penalty can be applied to the regions that surround all or part of the query molecule to prioritize hits without bulky extensions in constrained areas.RIDE APPLICATIONS:
  • Virtual screening (Efficiently screens “giga-sized” libraries)
  • Scaffold hoping (Identify structurally novel compounds based on the Reference)
  • Hit follow-up (Identifying new chemicals with similar 3D pharmacophoric properties to lead).

RIDE is built upon Atomic Property Fields (APF) technology, introduced by Max Totrov (2008 ref1), which is a continuous 3D molecular representation technique designed to encode and compare pharmacophoric and physicochemical characteristics of small molecules in a conformationally and spatially aware manner.

In APF, each heavy atom in a molecule contributes to seven atomic property channels:

1. 1. Hydrogen bond donor potential
   2. Hydrogen bond acceptor potential
   3. Sp² hybridization (aromaticity/planarity indicator)
   4. Electrostatic potential (based on partial charges)
   5. Hydrophobicity (atomic lipophilicity)
   6. Atomic size (van der Waals radii)
   7. Total charge

(Giganti et al, 2010) ref2 Table 3 compares the early enrichment performance of four 3D molecular alignment-based virtual screening tools—Surflex-sim, ROCS, FlexS, and ICMsim—across 12 DUD-E targets. Early enrichment (% of known actives retrieved) was assessed at the top 2%, 1%, and 0% of ranked libraries. ICMsim outperformed all methods with the highest mean enrichment at 2% (54.48%) and 1% (36.49%), followed by ROCS (35.63%, 17.78%) and FlexS (31.54%, 15.55%). Surflex-sim showed the lowest performance (2%: 23.15%; 1%: 9.07%). All methods dropped sharply at 0%, with ICMsim still leading (9.03%)

On target-by-target basis: 

• ICM was consistent across diverse protein families, including TK (90.01%), THR (88.89%), and DHFR (86.83%).
• ROCS, FlexS excels with ER (79.59%, 56.85%) and THR (86.36%, 81.82%). 
• Surflex-sim underperforms relative to other methods, with a mean enrichment of 23.15% (2%), dropping sharply at 1% (9.07%) and 0% (0.57%).

Table 4 evaluates the early-stage (1%) and late-stage (10%) enrichment performance of five virtual screening tools—Surflex-sim, ROCS, FlexS, ICMsim, and DOCK—using the Directory of Useful Decoys – Enhanced (DUD-E) dataset

• 11 protein targets: ADA, CDK2, DHFR, ER, FXA, HIVRT, NA, P38, thrombin, TK, trypsin
• ICMsim excels at 10%: Strong late-stage enrichment; effective for broad virtual screening.
• ROCS and FlexS at 1%: Better early enrichment due to accurate shape/pharmacophore alignment.
• DOCK and Surflex-sim: Lower overall performance; likely limited by scoring or alignment.

Available Databases:

• MolSoft has pre-generated conformer databases for a number of libraries (see table below). A user can generate conformers for other libraries using GINGER.

The table below contains information about the databases available and distributed to customers depending on their licensing:

Scalable Screening Engine

• Effective handling and processing of the databases
• ~ 5GB/1 M compounds with 30 conformers (1*30=30 M conformers)
• Highly parallelizable on both CPU and GPU versions
• Customizable query (Prioritize per atom contribution, envelop penalty, excluded volume(receptor)

Evaluation Metrics of RIDE

Below are the metrics used for virtual screening results:

EF1% (Enrichment Factor at 1%) per Target Protein: This metric quantifies the proportion of known active compounds ranked within the top 1% of the screened library for each target. High EF1% values for targets such as CDK2, HIVRT, and Thrombin demonstrate RIDE’s effectiveness in early recognition of bioactive ligands. This metric is particularly indicative of early hit identification performance and target-specific enrichment capability.

ROC AUC per Protein Target: ROC AUC values provide a measure of the overall ranking ability of actives versus decoys for each protein. Scores approaching 1.0 reflect strong discriminatory power. RIDE exhibits high AUC values for several target classes, including kinases and nuclear hormone receptors, indicating robust global ranking performance across diverse protein families.

BEDROC (Boltzmann-Enhanced Discrimination of ROC, α = 20.0). BEDROC emphasizes early enrichment by applying an exponential weighting toward the top of the ranked list. High BEDROC values for targets such as FXa, COX2, and DHFR confirm RIDE’s ability to prioritize actives early in the screening process, which is critical for real-time virtual screening workflows.

RIE (Robust Initial Enhancement): RIE provides a stabilized early enrichment metric, less susceptible to the imbalance between actives and decoys. It complements EF1% by validating early performance trends, particularly for protein targets with a limited number of known actives.

Top-N Actives per Protein Target: This metric reports the number of known actives retrieved within the top 10, 50, or 100 compounds in the ranked list for each target. It provides a tangible measure of practical screening success and facilitates decision-making in real-world hit identification campaigns.

RIDE Speed Benchmarking:

Benchmark Database: Screening collection from (1.5 M compounds, 30 M conformations)

Addittional Read: Novel GPU engines for virtual screening of gigasized libraries identify inhibitors of challenging targets.

References:

1. Totrov, Maxim. “Atomic property fields: generalized 3D pharmacophoric potential for automated ligand superposition, pharmacophore elucidation and 3D QSAR.” Chemical biology & drug design 71.1 (2008): 15-27.
2. Giganti, David, et al. “Comparative evaluation of 3D virtual ligand screening methods: impact of the molecular alignment on enrichment.” Journal of chemical information and modeling 50.6 (2010): 992-1004.
3. Michael M. Mysinger et al. “Directory of Useful Decoys, Enhanced (DUD-E): Better Ligands and Decoys for Better Benchmarking Journal of Medicinal Chemistry 2012 55(14): 6582-94.