Frequently Asked Questions

Pauling.AI is a fully autonomous computational drug discovery platform. You describe your target and research goals in plain language, and our AI agents execute the entire workflow: structure preparation, virtual screening, molecular dynamics simulations, free energy analysis, and ranked hit lists.

Unlike traditional computational chemistry software that requires expert operators, Pauling.AI is designed to be used directly by medicinal chemists, CSOs, and research scientists without a computational background. You stay focused on the science; we handle the infrastructure and execution.

A typical engagement starts with a target structure and a question. You provide a PDB ID or upload your own structure, specify the binding site if known, and describe your goals (hit screening, hit optimization, selectivity analysis, etc.).

From there the platform handles preparation (protonation, solvation, force field assignment), large-scale docking against our compound libraries, MD simulations for the top candidates, and free energy scoring. You receive a ranked hit list with binding poses, simulation trajectories, and supporting data ready for experimental follow-up.

Most projects require no more than an initial conversation with our team to scope and configure the run. Results are delivered in 24-72 hours depending on the project size.

Yes. Pauling.AI is a US-based company running on Google Cloud infrastructure with enterprise-grade security. Your protein structures, compound libraries, and results are stored in isolated, access-controlled environments. We do not use customer data to train models or share data between customers.

We are happy to sign NDAs and MSAs before any data is shared. For enterprise customers, we also offer private deployment options.

Our physics-based pipeline (docking + MD + mmPBSA) consistently outperforms docking-only approaches in prospective hit rate benchmarks. Molecular dynamics simulations capture protein flexibility and solvent effects that are invisible to rigid docking, significantly reducing false positives.

That said, computational predictions are a guide to experiment, not a replacement for it. Our goal is to give your team the highest-quality shortlist so wet-lab resources are focused on the most promising candidates, not spent on broadly screening the full library.

We support small molecule ligands. Our default screening libraries include approximately 10 million in-stock compounds from Enamine and mCule, prioritized because in-stock compounds eliminate procurement delays and allow immediate experimental validation.

You can also bring your own compound library in SDF format. Custom catalogs, internal compound collections, and focused libraries are all supported. Additional vendors are being added on an ongoing basis.

For large molecules such as cyclodextrins and other compounds exceeding standard size limits, we support GNINA-based docking pipelines capable of handling molecules up to 500 atoms.

We support protein-small molecule interactions as the primary use case. This includes:

• Monomeric proteins
• Multimeric complexes (homo- and hetero-oligomers)
• Membrane proteins and membrane-embedded complexes
• Proteins up to ~500 residues (larger targets available on request)

Protein-protein interactions, antibodies, and PROTAC research are on our roadmap and will be announced soon.

Yes. You can provide a PDB ID to pull directly from the RCSB Protein Data Bank, or upload your own structure in PDB or CIF format. Homology models, AlphaFold structures, and experimentally derived structures are all accepted.

Amino acid sequence input is also supported, with AlphaFold-based structure prediction integrated into the preparation pipeline. This is useful when no experimental structure is available or when you need to model a specific variant.

Our pipeline combines three complementary methods:

• Virtual screening (molecular docking) -- rapid scoring of large compound libraries against a target binding site using physics-based force fields. Default scoring uses the Vina score; Vinardo is also available.
• Molecular dynamics (MD) simulations -- all-atom simulations run with GROMACS on GPU-accelerated cloud infrastructure. MD captures protein flexibility and solvent effects, validating binding stability for top docking candidates. Standard runs are up to 100ns.
• Free energy analysis (mmPBSA/dG) -- rigorous thermodynamic scoring of protein-ligand binding affinity, providing a final ranking that correlates well with experimental IC50 values.

The default docking score is the Vina score (AutoDock Vina), which provides a good balance of speed and accuracy for large-scale screening. We also support the Vinardo scoring function, which can be selected depending on target class and project requirements.

For the top candidates, docking scores are supplemented by mmPBSA free energy calculations from MD simulations, which provide significantly higher accuracy for final ranking and selection.

Yes, full preparation is handled automatically. This includes protonation state assignment, addition of missing residues and loops, solvation, counter-ion placement, and force field parameterization for both protein and ligand.

Structures are stored in our database in mol2, SDF, PDBQT, and PDB formats. For MD simulations, we parameterize small molecules using GAFF2 (General Amber Force Field) and proteins with AMBER or CHARMM force fields depending on system requirements.

For MD simulations requiring membrane environments, we support standard lipid bilayer compositions and can customize membrane composition based on target biology.

For a full hit screening campaign -- 10 million compounds docked, top 100 candidates validated with MD simulations and mmPBSA free energy analysis -- turnaround is approximately 24 hours on our standard infrastructure.

Larger targets, extended MD simulation lengths, or custom analysis may increase turnaround. We will scope this clearly before starting any project.

Hit optimization campaigns (starting from known hits and generating improved analogs) have similar turnaround times depending on the number of iterations requested.

Our standard hit screening package (10M compound docking + MD validation for top 100 candidates) is priced at $10,000. This includes all compute costs, structure preparation, and delivery of a full ranked hit list with binding poses and simulation data.

For ongoing research collaborations, academic institutions, and multi-target projects, we offer tailored pricing. Contact us to discuss your project.

At minimum, you need a protein target. This can be:

• A PDB ID (we fetch the structure directly)
• A PDB or CIF file upload
• An amino acid sequence (we model the structure via AlphaFold3)

For MD simulations, we also ask for pH and temperature in Kelvin, as these affect protonation states and simulation parameters and improve correlation with wet-lab assays.

If you have prior knowledge of the binding site, allosteric pockets, or known inhibitors, that information is useful but not required -- our agents can identify binding sites autonomously.