Skip to content

binding-sites.md

Latest commit

 

History

History
65 lines (49 loc) · 5.26 KB

binding-sites.md

File metadata and controls

65 lines (49 loc) · 5.26 KB

ProtPocket Binding Site Analysis Pipeline

This document outlines the complete architectural lifecycle triggered when a user requests binding site analysis via GET /complex/{id}/binding-sites. The system relies on a heavily customized pipeline intersecting fpocket geometric void detection, and AlphaFold Structural B-Factors (pLDDT approximations).

1. Request Intiation & Metadata Retrieval

Location: handlers/bindingsites.go (BindingSiteHandler)

  • The endpoint normalizes a UniProt ID (e.g. Q55DI5) to ensure clean cache hashes.
  • FetchComplexData (services/alphafold.go) is invoked:
    • Queries the AlphaFold EBI Search API directly (/api/search?q={id}&type=complex).
    • Iterates results to locate both a Monomer (IsComplex = false) and the native/latest Dimer (IsComplex = true) variations.
    • Extracts the exact EBI model .cif asset URLs and their unique tracking entity IDs (e.g., AF-0000000066503175).

2. Geometric Pocket Detection (fpocket)

Location: services/fpocket.go (RunFpocket) Because AlphaFold protein models are entirely mathematical coordinate mappings without predefined cavities, fpocket runs Voronoi tessellation logic to map void properties.

  • Concurrent Execution: BindingSiteHandler spins up parallel Go-Routines to process the Monomer and Dimer CIF files simultaneously against fpocket.
  • Download Fallback: The parser attempts to replace suffix .cif with .pdb. If EBI responds 404 (due to size constraints blocking PDB conversions for multimers), it dynamically falls back to native .cif which fpocket also parses effortlessly.
  • Isolated Runtimes: Standard snap configurations for fpocket cannot cleanly interoperate with local /tmp structures on Linux securely. Thus, temporary runtime containers (tmp/fpocket-XXXXX) are provisioned exclusively for each routine in the local PWD.
  • Output Parsing:
    • parseFpocketInfo: Ingests structure_info.txt, gathering overarching chemical indicators (Druggability Score, Volume, Hydrophobicity, Polarity).
    • parsePocketAtoms (The Secret Sauce): Explicitly processes the output chunked .pdb per pocket (pocket[N]_atm.pdb). Crucially, this determines if a pocket spans chains (yielding geometrical IsInterfacePocket booleans) and extracts B-factor columns (61-66) caching localized structural pLDDT scores directly parsed from AlphaFold's structural translation!

3. Rigidity Modeling (Disorder Delta computation)

Location: services/plddt.go & services/pocket_filter.go AlphaFold explicitly maps its atomic B-factor arrays to 0–100 percentile confidence metrics (pLDDT).

To calculate Δ pLDDT (representing rigidification stabilization forces inherently caused by dimerization interactions), the server MUST baseline against the native monomer.

  • handlers/bindingsites.go invokes FetchMonomerPLDDT(monomerId). EBI successfully hosts _confidence_v4.json arrays specifically for standardized single-chain UniProts.
  • FilterInterfacePockets loops geometrically mapped Native Dimer B-factors (pulled from step 2).
    • Calculates Dimer B-Factor - Monomer JSON Score for each explicitly shared residue.
    • Generates the heavily-watched avg_disorder_delta (A positive score signifies strong pharmacological potential, proving that the localized interaction forces physically stiffened the loop natively upon binding!).

4. Sub-Pocket Pharmacological Binding (ZINC15 mapping)

Location: services/fragments.go Once structural properties are assembled natively, small-molecule binders are curated.

  • A concurrent sync.WaitGroup fires across every discovered Monomer and Dimer pocket simultaneously.
  • Queries zinc15.docking.org for molecules under 500 Da and logP < 5, mapping to the specified volume geometries.
  • ZINC parses the output recursively and serializes top hits into the models.Fragment interface array.

5. Topological Comparison (Monomer vs Dimer Mapping)

Location: services/pocket_compare.go (ComparePockets) The ComparisonResult strictly calculates differences conceptually missed by traditional linear matching.

  • Conformational Emergent Targets: A pocket physically generated via topological single-chain folding alterations specifically present inside complex variations but nowhere geographically close on the baseline Monomer. (Searches coordinates via < 6.0 Å Euclidean radius bounding boxes).
  • True Interface Pockets: An Emergent geometry that structurally spans explicitly distinct Chain IDs (Chain A & Chain B intersections).
  • Metrics Consolidation: Calculate Average Property Shifts (Volume, Polarity) natively, computing the DDGI (Dimerization Druggability Gain Index).

6. Response Layer & Frontend Render

Location: app/src/components/complex/BindingSitesPanel.jsx & ComparisonTab.jsx

  • The system responds to the HTTP client containing serialized coordinates. React automatically unpacks pocket_mapping and arrays.
  • BindingSitesPanel handles client-side limits natively, parsing grids visually mapped to the custom 3D web-GL library (Mol*).
  • ComparisonTab dynamically renders recharts to trace out the "Golden Quadrant" of Druggable stabilization anomalies, overlaying UI elements dynamically indicating missing data fields gracefully.