Geometric Analysis - Usage Guide

Overview

This skill measures geometric properties of protein structures with Biopython Bio.PDB: interatomic distances and distance matrices, bond and dihedral angles (phi/psi/chi and the Ramachandran list), superposition and RMSD, per-residue deviation, center of mass, radius of gyration, and solvent accessible surface area (SASA). Its governing idea is that no single number means "similar": every comparison metric is a choice of question, and RMSD in particular is a property of two structures GIVEN a superposition and an atom selection, not an intrinsic property of the pair. Residue contacts, contact maps, and interface/buried-surface analysis now live in structural-biology/interface-analysis; cross-protein fold comparison lives in alignment/structural-alignment; Ramachandran-as-a-quality-gate lives in structural-biology/structure-validation.

Prerequisites

pip install biopython numpy

Quick Start

Tell your AI agent what you want to do: - "Measure the distance between residue 50 CA and residue 100 CA" - "Superimpose these two conformations on the CA core and report the RMSD" - "Show me where the two structures actually differ, per residue" - "Compute per-residue relative SASA and list the buried residues" - "Calculate phi/psi angles for every residue"

Example Prompts

Distances and Angles

"What is the distance between these two atoms, and build a CA-CA distance matrix for chain A?"

"Calculate the phi/psi angles for all residues in this structure."

"Compute the chi1 angles for the aromatic residues."

Superposition and RMSD

"Superimpose the mobile structure onto the reference and report the CA RMSD."

"The whole-molecule RMSD looks large - show me the per-residue deviation so I can see whether it is just a loop or a real difference."

"These are two conformational states of the same protein - fit the rigid core and report core versus mobile separately."

Surface and Shape

"Calculate the solvent accessible surface area and tell me the probe radius used."

"Which residues are buried versus exposed by relative SASA?"

"What is the radius of gyration and the center of mass of this domain?"

What the Agent Will Do

Parse the structure file(s) and select the requested atom set, filtering waters and heteroatoms by hetflag.
For distances/angles, operate on Atom objects (subtraction for distance) or Vector objects from get_vector() (for calc_angle/calc_dihedral).
For superposition, build an equal-length ordered atom correspondence, fit with Superimposer (SVD/Kabsch), and report .rms plus, when useful, the per-residue deviation that exposes outlier domination.
For SASA, run ShrakeRupley.compute at the requested level and, for burial, normalize to the Tien et al 2013 max-ASA scale rather than reporting a bare absolute area.
Return measurements with units and state the choices (atom selection, superposition, probe radius) that the number depends on.

Key Functions

Function	Purpose
`atom1 - atom2`	Distance between atoms
`calc_angle()`	Angle between 3 atoms (Vector inputs)
`calc_dihedral()`	Dihedral angle from 4 atoms (Vector inputs)
`PPBuilder().build_peptides()`	Peptides for the Ramachandran phi/psi list
`Superimposer`	Rigid-body superposition + RMSD (SVD/Kabsch)
`Bio.PDB.qcprot`	QCP superposition, faster in tight loops
`ShrakeRupley`	Solvent accessible surface area

Tips

RMSD is not intrinsic to a pair of structures - always state the atom selection (CA? core? all-atom?) and the superposition it was measured under.
A large global RMSD often hides a near-identical core plus one mobile loop or a hinge - use per-residue deviation to see it, and fit the rigid core separately when the biology is a domain motion.
Superimposer needs equal-length ordered atom lists; it does NOT solve correspondence. Sequence-different structures need a structure-based alignment first (alignment/structural-alignment).
To compare or rank across proteins or lengths, do not stretch RMSD - use TM-score (>0.5 = same fold) or lDDT (superposition-free); those live with alignment/structural-alignment and the AlphaFold confidence skills.
Use get_vector() for angle/dihedral calculations, not .coord.
A SASA number is meaningless without its probe radius (1.4A water default); absolute SASA is not portable across tools, so prefer relative SASA (Tien 2013 max-ASA) for burial.
Superimposer.apply and atom.transform mutate coordinates in place - copy the structure first if the originals are still needed.
Before calling two structures "different states", check the difference against B-factors, resolution, and any NMR ensemble spread; a 1-2A core RMSD is often within noise.

structure-io - Parse and write PDB/mmCIF structure files
structure-navigation - Walk chains, residues, atoms; handle altlocs and disordered residues
structure-modification - Transform coordinates and edit structures in place
structural-biology/interface-analysis - Residue contacts, contact maps, and buried-surface interface analysis
structural-biology/structure-validation - Ramachandran and omega/cis-peptide outliers as a quality gate
alignment/structural-alignment - Cross-protein fold comparison and correspondence (TM-align, Foldseek, DALI)