BMRB Validation Pipeline

In this tutorial, we demonstrate how to validate synthetic protein models by programmatically fetching experimental data from the Biological Magnetic Resonance Data Bank (BMRB).

Prerequisites

You will need the following packages:

pip install synth-pdb pynmrstar requests

1. Fetching BMRB Entry Metadata

The BMRBAPI module allows us to fetch peer-reviewed assignments for structural validation. Let's start with human Ubiquitin (BMRB Entry 6457).

from synth_pdb.bmrb_api import BMRBAPI

# Fetch entry metadata
metadata = BMRBAPI.get_entry_metadata("6457")
print(f"Title: {metadata.get('title')}")
print(f"Authors: {', '.join([a['first_name'] + ' ' + a['last_name'] for a in metadata.get('authors', [])])}")

2. Downloading Experimental Chemical Shifts

Chemical shifts are the "fingerprints" of a protein's structure. We can fetch the assigned shifts and compare them to our synthetic predictions.

# Fetch experimental shifts
shifts = BMRBAPI.fetch_chemical_shifts("6457")

# Check shifts for the first residue (Met 1)
if 1 in shifts:
    print(f"Experimental Shifts for Met 1: {shifts[1]}")

3. Retrieving Distance Restraints (NOEs)

Distance restraints define the 3D fold. We can fetch the ground-truth upper bounds used by the original researchers.

# Fetch experimental distance restraints
restraints = BMRBAPI.fetch_restraints("6457")
print(f"Retrieved {len(restraints)} experimental NOE restraints.")

# Example: Look at the first restraint
if restraints:
    r = restraints[0]
    print(f"Restraint between {r['atom_name_1']}{r['index_1']} and {r['atom_name_2']}{r['index_2']}: Max {r['upper_limit']} Å")

4. Comparing Synthetic vs. Experimental Data

Now we generate a synthetic Ubiquitin structure and check if it satisfies the experimental NOE restraints.

from synth_pdb.generator import PeptideGenerator
from synth_pdb.analysis import GeometryAnalyzer

# 1. Generate synthetic structure (Human Ubiquitin sequence)
ubiq_seq = "MQIFVKTLTGKTITLEVEPSDTIENVKAKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQKESTLHLVLRLRGG"
gen = PeptideGenerator(ubiq_seq)
structure = gen.generate(conformation="alpha") # Start with a simple alpha helix

# 2. Check a restraint (e.g., a known long-range contact)
# (In a real pipeline, we would loop over all fetched restraints)
target_restraint = [r for r in restraints if abs(r['index_1'] - r['index_2']) > 10][0]
print(f"\nChecking restraint: {target_restraint}")

5. Benchmarking Geometric Quality

Finally, we can use the PDBValidationAPI to see how well-known experimental structures (like 1D3Z) rank in terms of geometric quality.

from synth_pdb.bmrb_api import PDBValidationAPI

# Fetch PDBe validation summary for 1D3Z
validation = PDBValidationAPI.get_validation_summary("1D3Z")
if validation:
    # Percentiles tell us how the structure ranks globally (higher is better)
    clash_percentile = validation[0].get("absolute_percentile_clashscore")
    rama_percentile = validation[0].get("absolute_percentile_ramachandran")

    print(f"1D3Z Global Clashscore Percentile: {clash_percentile}")
    print(f"1D3Z Global Ramachandran Percentile: {rama_percentile}")

Summary

By combining bmrb_api with synth-pdb, you can: - Benchmark your prediction algorithms against peer-reviewed data. - Validate that synthetic ensembles are biologically plausible. - Test structure calculation protocols with "perfectly assigned" experimental data.