LM-VS™ screens even for the most challenging targets — GPCR, MG, TPD, etc.
GPCR Target
We identify anchor residues within Open-Close membrane protein pockets to stabilize the ligand scaffold and initiate virtual screening.
Molecular Glue
We search for a molecular glue that acts as an adhesive between two weakly interacting proteins, thereby strengthening their interaction.
Target Protein Degrader
We identify the optimal compound that binds to the two pockets present in the degrader and the target, and then design a linker using the scaffolds of both compounds.
The entire process can be automatically executed
— on the STB Cloud
Planning
>
IV
>
LM-VS™
>
PAV
IV — Scaffold Design by MD simulation
Among the 10 MD snapshots, 1,4-difluorobenzene showed the least movement and was therefore selected as the scaffold, which is highlighted in red in Figure 1.
In Figure 2, R2 ring cluster indicates that the red-colored ring region interacts most frequently with the protein residues.
LM-VS™ — 10 Billion Compound Screening
By representing compound bonds within the pocket as 1D word strings and performing a Google-style inverted indexing search across a 10-billion-compound library, ultra-fast retrieval (as fast as 1/7000 second) becomes possible.
Compounds selected based on 1D string similarity are then re-docked, and those filtered by binding energy and scaffold similarity are defined as hit candidates.
In each iteration, the top 200 candidates are used for subsequent searches, and by repeating this process for 50 to 100 cycles, it becomes possible to screen scaffold-derived analogs that are synthetically accessible or commercially available.
PAV – View*Summary*Ranking
Post-Anal-View
This result visualizes the docking within the protein pocket, where the anchor residues (in yellow) interacting with the compound scaffold are fixed in position. The remaining portions of the compound are displayed as the screened derivatives, as shown in the figure.
Summary Report
(A) Hit Candidates Distribution — A 3D scatter plot illustrates the distribution of hit candidates according to CNNaffinity, CNNscore, and Ring Match Score.
(B) Accumulation of Hit Counts per Run — The cumulative number of strong and moderate hits is shown as the runs progress.
Excel Report & Function Rank
This report includes vendor information for compound synthesis and provides a functional score calculated from affinity parameters (CNNscore, CNNaffinity) and ADME descriptors (MW, LogP, HBA, HBD, MolMR, Solubility, Caco2 permeability, Rotatable Bonds, Strain Energy) according to the following formula.
AI redefines the pace of discovery.
Curious how it works?
Explore LM-VS™ — from input to insight.
Speed. Reliability. Scale.
Our 5MW AI Bio Supercom Center runs 5,000+ GPUs in parallel,
screening 10 billion compounds and generating results instantly.
5MW AI Bio Supercom Center
5,000+ CPU/GPU clusters
Massively parallel high-performance cores
4kW power + 20Mbps dedicated bandwidth per node
Independent cluster architecture for real-time stability
Instant results with zero data loss
AI efficiency redefines discovery.
Complete hit discovery for just $20K.
Timeline
Within 3 weeks
AI Learning Loops
Up to 50 Iterative Optimization
Compound Library
10 Billion (Enamine, Synple, SAVI & ZINC)
Deliverables
Top 200 Candidates Report + Structure Dataset
Turn your Hit into a Lead — for free.
Submit your SMILES string, and our AI will analyze its scaffold
to identify the most promising candidates from a 10B compound library.
Receive up to 200 lead molecules in your inbox — at no cost.
