Drug Discovery at Superluminal Speeds: Accelerating GPCR Programs with AI, Structure & High-Throughput Integration

Superluminal Medicines is pushing the boundaries of small-molecule drug discovery by combining protein structure, machine learning, and high-throughput experimentation. Their Hyperloop platform, used alongside LiveDesign by Schrödinger and CDD Vault, enables rapid virtual screening, conformational sampling, and hit-to-lead optimization. This blog walks through how they do it, what gains they’ve achieved, and what lessons you might apply.

Key Innovations at Superluminal

Dynamics of Protein Structure

• Superluminal emphasizes protein motions, not only static structures. They generate ensembles of conformations rather than relying on the most stable or observable state.
• Use of generative AI and modeling to map coordinate densities, simulate bond angles/lengths, and explore potential functional conformations.

Massive Virtual Libraries & Parallel Screening

• They leverage large enumerated libraries (tens of billions of compounds) and screen them in parallel across multiple conformations.
• Use computational filters to reduce candidates (ADMET, druglikeness, liability etc.), reducing synthetic overhead.

Integrated Structural Biology Infrastructure

• Operate decentralised cryo-EM (and/or microscopy) capabilities: 13 microscopes across six sites contribute to structure generation.
• Automated and ML-based imaging/data acquisition increases throughput ~5× over manual methods.

Conformation-Specific Targeting of GPCRs

• GPCRs are a focus, particularly MC4R. They identified specific conformations that yield biased signaling toward desired pharmacology.
• Use generative design (de novo) to generate compounds that are conformation specific, then validate in screening assays.

Speed with Quality

• For six GPCRs (including class B, which are challenging), Superluminal achieved hit-to-lead in less than five months per target.
• Their first development candidate is declared; remaining programs advancing.

Workflow Cohesion: LiveDesign & CDD Vault Integration

• Schrödinger’s LiveDesign provides a platform combining physics-based and ML approaches; enabling real-time feedback and design decision making.
• CDD Vault serves as the “single source of truth,” housing assay data, molecule registration, metadata, visualization, protocol templates, and integration via API for automation.

Use Cases & Measurable Gains

• Hit-to-lead timelines drastically reduced (GPCRs class B in under five months).

• Structural coverage improved: e.g., MC4R conformations that were not previously solved, now used for lead design.

• Throughput gains in structure determination (~5× more data acquisition with ML & decentralised cryo) vs manual.

Lessons & Considerations

• High computation or ML investment must be rooted in biology/chemistry questions—not just tools for their own sake.

• Parallel screening of many protein states increases chances of finding functionally selective ligands.

• Strong infrastructure (microscopes, experimental platforms) is critical; virtual models need experimental validation.

• Data management & workflow integration (like with CDD Vault + LiveDesign) greatly accelerate design, feedback, and decision-making.