Skip to content

Architecture

This document describes the high-level architecture of synth-pdb, a high-performance protein structure generator.

Design Philosophy

synth-pdb is designed to be a modular, extensible, and high-performance tool for generating synthetic protein structures. It follows a "Physics-First" approach, where structures are built using fundamental geometric and biophysical principles rather than relying on structural templates.

Core Components

The project is organized into several key modules, each responsible for a specific aspect of protein generation and analysis:

1. Structure Generation (synth_pdb/generator.py)

This is the heart of the tool. It handles: - Sequence Resolution: Determining the amino acid sequence from user input (length, specific sequence, or plausible frequencies). - Backbone Construction: Using the NeRF (Natural Extension Reference Frame) algorithm to build the N-CA-C backbone chain from internal coordinates (bond lengths, angles, and dihedrals). - Side-Chain Placement: Adding side-chain atoms based on rotamer libraries and backbone-dependent distributions.

2. Geometry & Physics (synth_pdb/geometry.py, synth_pdb/physics.py)

  • Geometry: Provides low-level mathematical functions for 3D transformations, dihedral angle calculations, and internal-to-Cartesian coordinate conversion.
  • Physics Engine: Implements energy functions (clash detection, electrostatic interactions, hydrogen bonding) and optimization algorithms (energy minimization, simulated annealing) to refine structures.

3. NMR & Biophysics (synth_pdb/nmr.py, synth_pdb/chemical_shifts.py, synth_pdb/rdc.py, synth_pdb/relaxation.py)

These modules bridge the gap between static structures and experimental observables: - Chemical Shifts: Predicts backbone and side-chain chemical shifts (supporting multiple predictors like SHIFTX2). - RDCs: Calculates Residual Dipolar Couplings for different alignment tensors. - Relaxation: Predicts NMR relaxation parameters (S², T1, T2, NOE) based on structural dynamics and tumbling.

4. Evolution & MSA (synth_pdb/evolution.py, synth_pdb/msa.py)

  • Simulates protein evolution to generate Multiple Sequence Alignments (MSAs).
  • Models co-evolutionary constraints to create realistic sequence variations that maintain structural integrity.

5. Validation (synth_pdb/validator.py)

Ensures that generated structures are biophysically plausible by checking: - Steric clashes. - Bond lengths and angles. - Ramachandran distribution. - Chirality.

Data Flow

A typical structure generation workflow follows these steps:

  1. Input Parsing: synth_pdb/main.py parses CLI arguments.
  2. Sequence Generation: A sequence is generated or provided.
  3. Backbone Building: generator.py uses geometry.py to build the backbone based on secondary structure presets (e.g., alpha-helix, beta-sheet).
  4. Side-Chain Addition: Rotamers are added to the backbone.
  5. Refinement: physics.py and packing.py optimize the structure to resolve clashes and minimize energy.
  6. Observables Calculation: If requested, NMR observables (shifts, RDCs, etc.) are calculated.
  7. Output: The final structure is saved as a PDB file, and metadata is exported to JSON or NEF formats.

Extensibility

The modular design allows researchers to easily plug in new: - Rotamer Libraries: By updating synth_pdb/data.py. - Force Fields: By extending synth_pdb/physics.py. - NMR Predictors: By adding new modules that implement the predictor interface.