Fragmenstein

The fragmenstein command is the main entry point for the CLI. It provides several high-level subcommands to manage different stages of the molecular stitching and combinatorial workflow:

• utils: Utilities for file manipulation, such as minimizing PDB files.
• monster: Stitching compounds (combining or placing) regardless of a protein context to create 'monster molecules'.
• victor: Stitching compounds (combining or placing) within a specific protein template.
• laboratory: Performs combinatorial operations.
• pipeline: Executes the full end-to-end pipeline.

Configuration can be adjusted via command-line arguments or environment variables.

fragmenstein [subcommand] [subcommand-options]

The mRMSD submodule provides a multiple RMSD variant that calculates a weighted '2-mean' of RMSDs across multiple hits. Unlike standard RMSD, this implementation does not perform superposition (alignment) and selects atoms based on a provided pairing (by default, atoms that were donors in Fragmenstein).

Requirements:

• RDKit is required.
• pyrosetta is not required.
• This submodule can be used independently of the Fragmenstein core functionality.

The Web UI provides a graphical interface for the following molecular workflows:

1. Upload: Import template proteins (PDB format) and hit fragment molecules (SDF, MOL, or PDB formats).
2. Combine: Perform pairwise merging to combine fragments into new ligand candidates.
3. Find analogs: Search for analogs using SmallWorld, PubChem, SMILES strings, or by uploading compound libraries (CSV/Excel).
4. Place: Position analogs within the protein binding site.
5. Browse results: Use interactive 3D visualization, view outcome charts, and export data via CSV or SDF.

Fragmenstein is a tool for merging, linking, and placing compounds by stitching bound compounds together. It performs two primary tasks:

1. Combine: Merges and links hits based on their atomic overlap.
2. Place: Places a given follow-up molecule based on one or more parent hits.

Note on Placement vs. Docking: Unlike traditional docking which uses pre-generated conformers to find a matching pose, Fragmenstein creates a 'monstrous conformer' from the parent(s) and then minimises it (optionally within a protein). This makes it a 'placement' tool rather than a standard docking tool.

The Victor class is the primary overarching class for molecular modelling in Fragmenstein. It coordinates several components, including Monster (for calculations) and Igor (for protein neighbourhood awareness).

Victor follows a specific workflow:

1. Receives a follow-up test, molecular objects of the inspiration, and a PDB template file.
2. Calls the Monster class.
3. Parameterizes the molecule.
4. Generates constraints.
5. Calls Igor.
6. Executes optional extra steps.

There are two main calculation methods:

• combine: Used for specific molecular combinations.
• place: Used to place a molecule into a site.

Unlike the Monster class, Victor's calculations are dependent on the protein neighbourhood due to its integration with Igor.

To avoid ambiguity in version 0.6.x and later, the following terminology is used:

• Blending: The process where hits are partially combined for placement purposes.
• Plonk: The placement of the final, distorted 3D 'monster' compound (referred to as 'Victor').
• Place/Position: Used synonymously.
• Combine/Merge: Used synonymously.

Fragmenstein supports different engines for energy minimization via the Victor class. You can choose based on your available hardware and required accuracy:

• PyRosetta (Default): Accessed via the Victor class. Uses Igor for minimization.
• OpenMM: Accessed via the OpenVictor class. Uses Fritz for minimization (requires a GPU).
• RDKit: Accessed via the Wictor class. Performs RDKit minimization within a frozen pocket (fastest, ~20s per compound).
• Quicktor: A variant tailored for speed.

Note: Igor (PyRosetta) takes approximately 60s per compound on a single core.

Results can be ranked using a ranker function that combines several metrics. A common implementation uses:

• LE (Ligand Efficiency)
• comRMSD (Combined RMSD)
• atom_bonus (based on the number of constrained atoms)
• novelty_penalty (ratio of unconstrained to constrained atoms)

Example weights: rank_weights = {'LE': 1., 'comRMSD': 2., 'atom_bonus': 2. , 'novelty_penalty': 5.}

The ring merger in Fragmenstein is not transitive. The order in which molecules are passed to the Monster constructor affects the resulting conformation of the positioned_mol. For example, merging [ortho, meta] may produce a different (potentially more distorted) conformation than merging [meta, ortho] for the same compound (e.g., methylcatecol).

# Example of hit order affecting conformation
Monster([ortho, meta]).merge().positioned_mol  # decent
Monster([meta, ortho]).merge().positioned_mol  # indecent

Fragmenstein applies several chemical constraints and corrections during the merging process:

• Warhead Harmonisation: Bonding directly to a warhead is forbidden. When merging hits with different warheads, the system may ignore, keep, or use the first warhead, which can lead to unexpected linkages.
• Fused Rings: To avoid unlikely chemical structures, bridges are removed, and rings of size 3 or 4 are corrected to size 5 (e.g., cyclopropane-benzene is corrected).
• Forbidden Structures: Allene structures are forbidden.
• Perpendicular Rings: If two rings intersect perpendicularly, the system may issue an "emergency bonding" warning and produce unexpected bonding.
• Spiro Compounds: These are explicitly tolerated.