MagHelix™ CADD Platform
Deep Learning Virtual Screening
Molecular Dynamics & Free Energy
De Novo Design & ADMET
Traditional HTS delivers 0.01–0.1% hit rates. Our AI-driven CADD platform achieves 5–20% on 716-core HPC — from small molecules to antibodies — with direct handoff to Structural Biology and Lead Optimization.
Creative Biostructure at a Glance
Over a decade of trusted expertise powering biotech, pharma, and research institutions worldwide to advance therapeutic innovation.
Why Partner With Us
Most virtual biotechs cannot justify the $500K+ annual license stack for Schrödinger, MD simulation suites, and compound databases — let alone the GPU cluster required to run them at scale. Pharma teams often split docking, MD, and ADMET prediction across separate vendors, losing the structural context that links a binding pose to a toxicity alert. We built this platform to eliminate that fragmentation: one HPC infrastructure where AI scoring, physics-based simulation, and generative design share the same compound database and project ID.
Your CapEx is in chemistry and biology. Ours is in software licenses, GPU clusters, and compound libraries.
| Stage | What We Deliver | What You Don't Need to Build |
|---|---|---|
| Target Assessment | AlphaFold3 structure prediction; homology modeling; pocket druggability scoring | Structural bioinformatics team |
| Hit Identification | AI-enhanced virtual screening (SBVS/LBVS); molecular docking; pharmacophore matching; antibody humanization & affinity maturation | Software licenses, compound libraries |
| Hit Validation | Molecular dynamics stability analysis; FEP affinity ranking; biophysical validation design | GPU cluster, MD expertise |
| Lead Optimization | De novo design; fragment-to-lead growing; QSAR modeling; ADMET prediction | Medicinal chemistry informatics |
| Data Handoff | Docking-ready coordinates + optimized compound series with binding mode rationale | — |
Production-Ready Deliverables: Every campaign ships with ranked compound lists, docking poses, MD stability reports, FEP affinity estimates, and ADMET risk profiles — ready for synthesis prioritization or X-ray co-crystallization.
- ✓ Milestone-based pricing aligned with your fundraising cycles
- ✓ No software license overhead — Schrödinger, DOCK, Modeller, ChemAxon, and CCDC under one roof
Allosteric pockets. PPI interfaces. Undruggable kinases. Antibody humanization. "Undruggable" is our starting point.
Proven platform scale
58-blade, 716-core HPC cluster screening 1 million+ compounds in 10 days; 2 million+ unique compound database (ZINC, MDDR, ACD, NCI) plus in-house Drug Information system.
Small molecules to biologics
AI-driven antibody humanization, affinity maturation, and peptide optimization alongside traditional small-molecule CADD.
IP firewall & encrypted data infrastructure
Full audit trails, client-isolated project folders, and GLP-ready documentation.
Core Service Modules
Service Module At-a-Glance
| Service | Core Capability | Platform Integration | Typical Timeline |
|---|---|---|---|
| AI-Enhanced Virtual Screening | SBVS & LBVS; deep learning docking (GNINA, DiffDock); pharmacophore modeling; 1M+ compounds/10 days | Direct handoff to Molecular Docking and Structural Biology for validation | 1–2 weeks |
| Molecular Dynamics & Free Energy | All-atom MD; induced-fit docking; FEP/TI; MM/PBSA affinity rescoring | AlphaFold3 ensemble input; Cryo-EM density validation; direct link to Lead Optimization | 2–4 weeks |
| De Novo Design & Fragment-to-Lead | Generative AI scaffold design; pocket druggability; hot-spot analysis; SAR-by-catalog; QM refinement | Integration with FBDD and Hit-to-Lead pipelines | 2–6 weeks |
| ADMET Prediction & Target Profiling | hERG, P450, BBB permeability, bioavailability prediction; reverse docking target fishing; toxicity alert scoring | Direct handoff to In Vitro ADME-Tox and Zebrafish Screening for experimental validation | 1–2 weeks |
AI-Enhanced Virtual Screening
From Million-Compound Libraries to Validated Hits
Key Features:
- Structure-Based Virtual Screening (SBVS) — Molecular docking with AutoDock Vina, Glide, and GNINA deep learning scoring; pharmacophore modeling based on receptor pockets or ligand features.
- Ligand-Based Virtual Screening (LBVS) — Similarity searching, 3D-pharmacophore search, and QSAR modeling to identify novel scaffolds from known actives.
- AI Scoring Functions — GNINA convolutional neural networks and DiffDock diffusion models outperform traditional force-field scoring in pose prediction and virtual screening enrichment.
- Platform Throughput — 716-core HPC cluster screens >1 million compounds in 10 days from our 2 million+ compound database (ZINC, MDDR, ACD, NCI) plus in-house Drug Information system.
What We Offer: For virtual biotechs, this eliminates the need to procure Schrödinger licenses, curate compound libraries, or maintain GPU clusters. For pharma, our AI-enhanced scoring reduces false positives by 40–60% compared to traditional docking, conserving synthesis capacity for high-confidence hits.
Molecular Dynamics & Free Energy Simulation
Physics-Based Validation of AI Predictions
Key Features:
- All-Atom MD Simulation — GROMACS, AMBER, and Schrödinger Desmond for protein-ligand complex stability analysis in explicit solvent and lipid bilayers.
- Induced-Fit Docking — Receptor flexibility modeling capturing ligand-induced conformational changes invisible to rigid-body docking.
- Free Energy Perturbation (FEP/TI) — FEP and MM/PBSA rescoring for lead optimization affinity ranking and selectivity profiling.
- Structural Validation — MD trajectories validated against experimental Cryo-EM densities and X-ray B-factors; AlphaFold3 ensembles provide starting conformations for flexible targets.
What We Offer: For hit-to-lead programs, FEP-guided analog ranking predicts affinity changes before synthesis, reducing design cycles by 30–50%. For difficult targets like membrane proteins, MD in explicit lipid bilayers reveals binding site accessibility that static structures miss.
De Novo Design & Fragment-to-Lead
Generative AI for Novel Scaffolds and Fragment Growing
Key Features:
- De Novo Drug Design — Generative models (RNNs, VAEs, reinforcement learning) create novel molecular structures with desired pharmacological properties from target structure or ligand pharmacophores.
- Fragment-to-Lead (F2L) — Pocket druggability prediction, hot-spot analysis, and SAR-by-catalog strategies grow fragment hits into lead-like compounds with synthesizable exit vectors — directly integrated with our FBDD Platform.
- Quantum Mechanics (QM) — Ab initio calculations and QM/MM hybrid methods refine binding mechanism details, particularly for covalent inhibitors and metal-coordinating ligands.
- Scaffold Hopping — 3D-pharmacophore search and AI generative models escape patent constraints while preserving target affinity.
What We Offer: For fragment-based screening programs, our F2L pipeline transforms STD-NMR or X-ray fragment hits into optimized leads. For novel targets, de novo design generates patentable scaffolds with predicted ADMET compliance.
ADMET Prediction & Target Profiling
Predicting Liability Before Synthesis
Key Features:
- ADMET Modeling — Machine learning models predict hERG cardiotoxicity, CYP450 inhibition, blood-brain barrier permeability, P-gp substrate liability, and oral bioavailability — trained on curated datasets from our in-house Drug Information system.
- Target Fishing — Reverse docking and pharmacophore searching identify off-target liabilities and polypharmacology profiles for existing actives.
- Toxicity Alert Scoring — Structural alerts for mutagenicity, carcinogenicity, and hepatotoxicity flagged before compound synthesis.
- Experimental Validation Loop — Predictions cross-validated with In Vitro ADME-Tox and Zebrafish Screening to refine model accuracy.
What We Offer: For lead optimization teams, ADMET predictions prioritize analogs with favorable safety profiles before committing to expensive synthesis. For repurposing campaigns, target fishing reveals new indications or toxicity liabilities for known drugs.
Technology Platform
Integrated CADD Infrastructure: AI Scoring + Physics Simulation + Generative Design + ADMET Prediction, Zero Handoffs
Traditional CADD splits virtual screening, MD validation, and ADMET analysis across separate software stacks and vendors. Our platform unifies all stages on shared HPC infrastructure, with AI predictions informing MD simulations and experimental validation data feeding back into model retraining.
Computational Platform — Dry Lab
| Capability | Details |
|---|---|
| AI Docking & Scoring | GNINA (CNN scoring), DiffDock (diffusion-based pose generation), AutoDock Vina, Glide; geometric deep learning for affinity prediction |
| Molecular Dynamics | GROMACS, AMBER, Schrödinger Desmond; explicit solvent and membrane simulations; enhanced sampling (metadynamics, replica exchange) |
| Free Energy Calculations | Schrödinger FEP+, AMBER TI, MM/PBSA; lead optimization affinity and selectivity ranking |
| Generative Design | ReLeaSE (RNN + RL), VAE-based scaffold generation, pharmacophore-constrained de novo design |
| Fragment-to-Lead | Pocket druggability prediction; hot-spot analysis; SAR-by-catalog; QM/MM binding mechanism refinement |
| Data Fusion & ML | Multi-algorithm consensus scoring combining docking, pharmacophore, and QSAR predictions for robust hit prioritization |
| QSAR & ADMET | CoMFA, CoMSIA, machine learning classifiers (Random Forest, XGBoost, GNN) for hERG, CYP450, BBB, solubility prediction |
| Data Infrastructure | Client-isolated project folders; full audit trails; automated pipeline logging from docking to MD |
Hardware & Database Resources
| Resource | Specification |
|---|---|
| HPC Cluster | 58 blades, 716 cores; GPU acceleration (A100/H100) for deep learning docking and generative models |
| Throughput | >1 million compounds screened in 10 days; parallel MD simulations for 100+ ligands simultaneously |
| Compound Databases | 2 million+ unique compounds: ZINC, MDDR, ACD, NCI, plus proprietary in-house Drug Information system |
| Software Stack | Schrödinger Drug Discovery Suite, DOCK, Modeller, ChemAxon JChem Suite, CCDC, PyMOL, ChimeraX |
| AI/ML Frameworks | PyTorch, TensorFlow, RDKit, Open Babel, scikit-learn for custom model development |
Platform Edge: The ability to screen 1 million compounds with AI docking on Monday, run FEP affinity ranking on the top 500 hits by Wednesday, and deliver a ranked compound list with ADMET risk profiles by Friday — all on shared infrastructure with unified project tracking.
Shimadzu LC-20AP Prep-HPLC
Thermo Fisher Krios G4
PerkinElmer Operetta CLS
Bruker Avance NEO 800 MHz
Platform specifications are subject to continuous upgrade. Contact our team for instrument availability and project-specific capability assessment.
Closed-Loop Discovery Engine
When Computation Meets Experiment
Static docking scores predict binding poses. Experimental structures and biophysical data reveal the reality that models miss — induced-fit changes, entropy penalties, and off-target liabilities. Our platform feeds every experimental result back into the design cycle.
AI Virtual Screening
GNINA/DiffDock screen millions of compounds; AlphaFold3 structures provide receptor conformations for novel targets
→ Feeds into MD Validation
Physics-Based Validation
MD and FEP refine docking poses; binding free energy estimates prioritize synthesis candidates; Cryo-EM and X-ray structures validate predicted modes
→ Feeds into ADMET Filtering
ADMET & Target Profiling
ML models predict hERG, CYP450, BBB liability; reverse docking flags off-target risks; only compliant compounds progress to synthesis
→ Feeds into Experimental Testing
Industrial Value:
For Biotechs
Your first campaign's docking and MD data trains the AI models for your second target. Experimental validation from Phase 0 becomes training data for Phase 1 — a compounding learning partnership.
For Pharma
Every computational prediction is linked to an experimental outcome with project ID, timestamp, and model version — fully audit-ready for regulatory submissions and internal portfolio reviews.
Project Management & Execution
Project Workflow
A standardized, milestone-driven execution system. From target structure to ranked compound series.
01 Target Assessment
- AlphaFold3 structure prediction; pocket druggability analysis; homology modeling if no experimental structure
Deliverable: Target structure report with pocket analysis
02 Virtual Screening
- AI-enhanced docking (GNINA/DiffDock); pharmacophore screening; 1M+ compound library filtering; hit ranking by CNN score
Deliverable: Ranked hit list (top 1–5%) with binding mode rationale
03 MD & Free Energy Validation
- All-atom MD stability analysis; FEP/TI affinity ranking for top 100–500 hits; induced-fit validation
Deliverable: MD trajectories + FEP affinity ranking + selectivity profile
04 ADMET Filtering
- hERG, CYP450, BBB, solubility prediction; target fishing for off-target liability; toxicity alert filtering
Deliverable: ADMET risk assessment + target liability report
05 Deliverables & Handoff
- Ranked compound list with docking poses; MD stability report; FEP affinity estimates; ADMET risk profile; handoff to SBDD or Lead Optimization
Deliverable: Final compound package + synthesis priority recommendation + transition plan
Sample Requirements
| Sample Type | Specification |
|---|---|
| Target Structure | PDB ID; AlphaFold3 model; or FASTA sequence for homology modeling |
| Known Ligands | Reference compounds with activity data (if available) for pharmacophore/QSAR model training |
| Compound Library | Custom SDF/MOL2 files welcome; otherwise screened against our 2M+ database (ZINC, MDDR, ACD, NCI) |
| Screening Scope | Full library vs. focused subset; rigid vs. flexible docking; inclusion of covalent or allosteric sites |
| Prioritization Criteria | Affinity threshold; ADMET cutoffs; synthetic accessibility; patent landscape constraints |
Standard Deliverables
- Ranked compound list (top 1–5%) with docking poses and CNN confidence scores
- Molecular dynamics stability report for top-ranked complexes
- FEP or MM/PBSA affinity and selectivity estimates
- ADMET risk profile (hERG, CYP450, BBB, solubility, toxicity alerts)
- Target fishing report for off-target liabilities
- Direct handoff to Molecular Docking, Structural Biology, or Lead Optimization
Our technical team responds within 24 hours. All inquiries protected under NDA.
Frequently Asked Questions
A: Virtual screening: 1–2 weeks. MD validation + FEP: 2–4 weeks. De novo design: 2–6 weeks. Full target-to-lead campaigns typically complete within 4–8 weeks with milestone gates at hit identification, MD validation, and ADMET filtering.
Case Study
Case: Structure-Based Virtual Screening against IQGAP1 GTPase-Activating Protein-Related Domain (GRD)
Goal: Identify small-molecule inhibitors targeting the GRD–Cdc42 protein–protein interaction interface via large-scale structure-based virtual screening — demonstrating the MagHelix™ CADD Platform's capacity to process >200,000 compounds and deliver ranked, cluster-analyzed hit lists with binding mode rationale.
Key Data:
- Findings: Two distinct pockets identified at the PPI interface —
Pocket A (flat, shallow) and Pocket B (druggable). Most top-ranked
compounds occupied Pocket B or A+B. Representative hits:
- AK-778/41507563 (rank 5, −7.98 kcal/mol): Pocket B, H-bonds to N1189/V1045, π–π with Y1192
- AK-968/41172133 (rank 41, −7.41 kcal/mol): Pocket A+B, H-bonds to R1194/K1053/N1189, π–π with Y1192
- AB-323/13887426 (rank top 100, −7.40 kcal/mol): Pocket B, multiple H-bonds and π–π interactions
- AS-871/43477026 (rank 54, −7.32 kcal/mol): Pocket B only, H-bonds to R1194/K1053/Y1193/N1197
Why it matters: This case demonstrates our platform's end-to-end SBDD workflow — from high-resolution crystal structure preparation and large-library docking (>200K compounds) to cluster-based hit prioritization and detailed interaction mapping. For virtual biotechs targeting PPIs, this eliminates the need to build in-house docking infrastructure. For pharma, the delivered package (PDB receptor file, top 1,000 clustered SDF/Excel, top 10,000 SDF, and star-rated visual inspection notes) is ready for immediate procurement and biophysical validation.
Figure 1. IQGAP1 GRD (pink) in complex with Cdc42 (yellow). Conserved ¹¹⁹²YYR¹¹⁹⁴ motif and key residues shown as sticks; Mg²⁺ as magenta sphere. Red arrow: virtual screening grid center at R1194.
Figure 2. Pocket analysis of the GRD-Cdc42 PPI interface. GRD surface in gray; Cdc42 in yellow cartoon. Pocket A and Pocket B identified; solvent-accessible residues Y1193 and R1194 highlighted in red.
Figure 3. Top-ranked compounds bound to GRD pockets. GRD shown as gray surface; compounds in orange stick. Left panels: 2D interaction maps with H-bonds (green dashed) and π–π stacking (orange dashed).
Need AI-accelerated CADD to power your drug discovery program? Our team can design a customized computational chemistry pipeline tailored to your target class, compound constraints, and regulatory milestones. Contact our scientific team today.
