PocketFlow

What is PocketFlow?

PocketFlow is a structure-based deep generative model that designs novel drug-like molecules inside protein binding pockets. Published in Nature Machine Intelligence in March 2024 by researchers at Sichuan University, it combines autoregressive flow modeling with explicit chemical knowledge to generate molecules with 100% chemical validity.

What sets PocketFlow apart is its experimental validation. The authors applied PocketFlow to design inhibitors for two epigenetic targets (HAT1 and YTHDC1) and successfully obtained wet-lab-validated bioactive lead compounds. This makes PocketFlow the first structure-based molecular deep generative model with experimental validation of designed molecules.

In computational benchmarks on the CrossDocked2020 dataset, PocketFlow outperforms previous methods while maintaining perfect chemical validity and high drug-likeness scores.

How does PocketFlow work?

PocketFlow generates molecules atom-by-atom within a protein binding pocket using an autoregressive approach. At each step, the model decides what atom type to add, where to place it in 3D space, and how to connect it to existing atoms. Chemical rules guide these decisions to ensure valid molecules.

GDBP architecture

The core of PocketFlow is the Geometric Double Bottleneck Perceptron (GDBP), an SE(3)-equivariant graph neural network that models the 3D geometry of the protein-ligand complex. GDBP improves upon earlier geometric neural networks (GVP and GBP) by adding bottleneck layers for both scalar and vector features.

This architecture processes 3D coordinates directly while maintaining equivariance to rotations and translations. The model can generate atom positions in 3D space without needing to first predict internal coordinates.

Generation components

The autoregressive generation uses three specialized components working together:

Atom Flow predicts the type of each new atom (carbon, nitrogen, oxygen, etc.) using a normalizing flow. This probabilistic approach captures the distribution of atom types conditioned on the current molecular state and pocket environment.

Position Predictor determines where to place each new atom in 3D space relative to the binding pocket. The GDBP network encodes spatial relationships between existing atoms, protein residues, and potential placement sites.

Bond Flow predicts connectivity between the new atom and existing atoms using another normalizing flow. This component receives explicit chemical knowledge guidance to ensure reasonable bond patterns.

Chemical knowledge integration

Unlike purely data-driven approaches, PocketFlow incorporates chemical knowledge directly into the generation process. The bond predictor checks whether proposed bonds satisfy valence rules and reasonable bonding patterns.

If the model proposes an unreasonable bond, it resamples until finding a valid connection. This explicit guidance is critical—ablation studies show that removing chemical constraints significantly degrades both validity and drug-likeness of generated molecules.

Training

PocketFlow uses a two-stage training process. The model is first pretrained on the ZINC 3D database of drug-like molecules to learn general molecular patterns. It is then fine-tuned on CrossDocked2020, a dataset of protein-ligand complexes, to learn pocket-specific generation.

Inputs & settings

Binding pocket

By default, provide a full protein structure in PDB format plus a positioned reference ligand pose in SDF or MOL format. ProteinIQ uses PocketFlow's native pocket-splitting helper to extract residues near the ligand before generation.

The reference ligand is not docked. Its 3D coordinates define the binding site, and native PocketFlow keeps nearby residues using the pocket cutoff setting. The native preprocessing default is 10 Å.

If you already have a PocketFlow-prepared pocket PDB, switch Pocket input to Pre-extracted pocket and upload it directly. Every ATOM record must end with PocketFlow's surf or inner annotation because the model uses that surface mask to choose its first focal atoms. An ordinary pocket PDB does not contain this information; use full-protein mode when you do not already have an annotated pocket.

Generation settings

• Number of molecules: How many molecules to generate (1-1000). More molecules provide greater chemical diversity but increase computation time. The default is PocketFlow's command-line default of 100 molecules.
• Maximum atoms: Upper limit on heavy atoms per molecule (15-50). The default is PocketFlow's command-line default of 40 heavy atoms. Smaller limits produce fragment-like compounds suitable for fragment-based drug design.
• Output mode: Interactive viewer mode returns every generated molecule as an individual SDF file for inspection. Batch library mode preserves aggregate outputs only and previews molecules directly from generated.sdf.
• Pocket cutoff: Distance cutoff for native pocket extraction from the reference ligand pose. Residues whose center of mass is within the cutoff are retained. The default is 10 Å.

Sampling parameters

These advanced settings control the stochastic generation process.

Temperature parameters affect sampling diversity. Lower temperatures produce more conservative, predictable structures while higher temperatures explore more unusual chemical space.

• Atom temperature: Controls randomness in atom type selection. Values below 1.0 favor common atom types, above 1.0 increases diversity.
• Bond temperature: Controls randomness in bond type selection. Lower values favor common bonding patterns.

Focus parameters control how the model selects which atom to extend next during generation.

• Focus threshold: Minimum focus probability used only when greedy focus selection is disabled. Higher values narrow the eligible focal candidates.
• Greedy focus selection: When enabled, PocketFlow picks the highest-probability focal candidate instead of applying the focus threshold. Disable it to sample from eligible candidates.
• Initial protein-ligand distance: Distance cutoff used when PocketFlow places the first generated atom relative to the protein pocket.

Understanding the results

PocketFlow returns generated molecules in native generation order. ProteinIQ also calculates QED (Quantitative Estimate of Drug-likeness) and standard molecular properties as derived annotations for easier triage.

PocketFlow does not predict binding affinity or activity. Use downstream docking tools such as AutoDock Vina or GNINA when you need binding-oriented ranking.

QED score

QED ranges from 0-1, with higher values indicating more drug-like properties. The score combines molecular weight, lipophilicity, hydrogen bond donors/acceptors, polar surface area, rotatable bonds, and aromatic rings into a single metric.

• QED > 0.7: Highly drug-like
• QED 0.5-0.7: Moderate drug-likeness
• QED < 0.5: Less drug-like, may require optimization

Molecular properties

• MW (Da): Molecular weight. Drug-like molecules typically fall in the 200-500 Da range.
• LogP: Predicted octanol-water partition coefficient. Values between 1-5 are typical for orally bioavailable drugs.
• SMILES: Canonical SMILES representation for each molecule.

Downloading results

Download individual molecules as SDF files for further analysis in molecular modeling software. The 3D coordinates correspond to the predicted binding pose within the pocket.

Batch library mode returns aggregate files for downstream screening:

• generated.sdf: Multi-record SDF containing all parsed generated molecules.
• generated.smi: SMILES list with one generated molecule per line.
• metrics.dir: PocketFlow's native generation metrics.
• readme.txt: Native command arguments and run description from the pinned PocketFlow entrypoint.
• molecules.csv: ProteinIQ-derived molecule manifest with generation order, SMILES, QED, and standard molecular properties.

The browser viewer reads individual molecule records from generated.sdf when you select a row, so batch library jobs do not create separate molecule_###.sdf preview files.

Best practices

Pocket preparation

Use a ligand pose already positioned in the intended binding site. If the ligand pose is misplaced, the extracted pocket will be misplaced too.

Extracted pockets need sufficient context (8-12Å from the binding site center). Too small pockets may constrain generation, while overly large pockets increase noise.

Remove crystallographic artifacts like buffer molecules, and consider whether to keep structural waters based on their role in binding.

Iterative design

Use PocketFlow as part of an iterative design workflow:

1. Generate an initial set of molecules with default settings
2. Analyze generated compounds for favorable derived properties
3. Adjust parameters (temperatures, atom limits) based on results
4. Validate promising candidates with docking tools like AutoDock Vina or GNINA
5. Assess ADMET properties with ADMET-AI

Limitations

• Full-protein mode requires a positioned reference ligand pose; PocketFlow does not dock the ligand
• Pre-extracted pockets require PocketFlow's per-atom surf/inner annotations
• Generated poses are approximate and benefit from refinement via docking
• Does not account for protein flexibility
• Some chemically valid molecules may still be synthetically challenging

Additional resources

• PocketFlow GitHub Repository
• Nature Machine Intelligence Paper

---