At Darien Biodiscovery, we have built a proprietary technology stack specifically engineered to tackle RNA’s unique challenges – from its dynamic structures to its elusive binding pockets. Our hybrid approach combines physics-based rigor with AI’s pattern recognition power, enabling us to deliver results where traditional methods fail.

Core Technology Pillars

RNA Spatial Structure Modeling

a. Physics-Based Foundations

• All-Atom Molecular Dynamics (MD): Simulate RNA folding & ligand binding at μs-ms timescales using specialized force fields (RNA-OL3, AMBER-DH2).

• Enhanced Sampling: Accelerate rare-event sampling with GaMD, MetaDynamics, and AI-driven collective variable discovery.

• Co-Transcriptional Folding Predictions: Model RNA structure formation in real-time as nucleotides are synthesized.

b. Deep Learning Revolution

• 3D Capsule Networks: Predict tertiary structures from sequence alone using our DeepFoldRNA™ model, trained on 1,500+ experimentally solved RNA structures.

• Ensemble Modeling: Generate probabilistic structure ensembles to capture RNA’s intrinsic flexibility.

RNA-Focused Virtual Screening

a. Gigascale Docking Infrastructure

• GPU-Accelerated Workflows: Screen 10B+ compounds in days using optimized AutoDock-GPU pipelines.

• Target-Adaptive Grids: Dynamically adjust docking grids to RNA’s flexible regions (loops, bulges).

b. Machine Learning Filters

• RNA-ChemSpace Navigator™: Prune non-RNA-compatible chemotypes using 15+ RNA-specific filters (charge, solvation, motif matching).

• Non-Enumerated Library Support: Screen ultra-large spaces via SMIRKS-based reaction rules & generative subgraph sampling.

Lead Optimization Engine

a. Free Energy Mastery

• Alchemical FEP: Calculate ΔΔG binding with <1 kcal/mol error using bespoke RNA solvent models.

• QM/MM Hybrid: Refine critical interactions (e.g., cation-π, halogen bonds) at the DFT level.

b. AI-Driven Property Prediction

• ADMET-RNA™: Predict tissue penetration, efflux ratios, and RNA-specific toxicity risks.

• Synthetic Accessibility Scoring: Prioritize compounds with 3-click MEDCHEM routes using our retrosynthesis-trained AI.

Proprietary Breakthrough: RiboAffinity

Our key technology pillar – the first AI scoring function trained exclusively on RNA-ligand dynamic interactions – redefines affinity prediction accuracy:

Why It’s State-of-the-Art:

• Largest Training Set: >250 experimental nucleic acid-ligand complexes augmented with template-based modeling and molecular dynamics data and binding data points (Kd, IC50, ΔG) across various RNA target classes.

• Multimodal Input: Processes 3D RNA-ligand poses, sequence context, and solvent dynamics.

• Outperforms Legacy Tools: 30% higher enrichment vs. Glide/AutoDock in our in-house RNA benchmark sets.

Key Applications:

• Rank-order virtual screening hits with experimental-level accuracy

• Explain binding via attention maps, highlighting critical RNA nucleotides

Technology Integrations

• Generative Chemistry: VAEs & diffusion models for RNA-targeted library expansion.

• CRISPR Design Suite: Optimize guide RNAs via in-house deep learning models (DeepGuide™).

• Cloud-Native Architecture: AWS-optimized pipelines for burst scaling to 128+ GPUs.

Why Our Technology Wins

• Hybrid AI/Physics Approach: Get the robustness of physics with AI’s speed.

• RNA-Specific Tuning: No more repurposed protein tools.

• Scalability Meets Precision: From gigascale screens to quantum-level refinements.