DHEA Target Protein Prediction
Integration of in silico and indirect experimental methods
Integrated Process
Goal of prediction phase: reduce from ~20,000 human proteins to ~10–20 top candidates for DARTS with targeted Western blot (with antibodies), saving time and cost before full proteome MS analysis
~20k → ~10–20 candidates
Database Lookup — Known Targets & Similarities
Before using prediction tools, look up directly in databases. DHEA is an important endogenous steroid with many protein interactions already recorded.
ChEMBL
FreeMost comprehensive bioactivity database. Find DHEA → view all assay activity with proteins. Filter by IC₅₀, Ki, Kd to identify strong targets.
BindingDB
FreeDirect measurement data of binding constants (Kd, Ki, IC₅₀) between ligand and protein.
DrugBank
Free/ProComplete pharmacological info for DHEA: known targets, pathways, drug interactions, mechanisms.
STITCH
FreeSimilar to STRING but for chemical–protein interactions. View DHEA interaction network with confidence scores.
PubChem BioAssay
FreeThousands of HTS bioassay results related to DHEA from NIH large-scale screening.
OMIM / DisGeNET
FreeLink diseases with genes/proteins. Proteins related to diseases DHEA treats → pathogenic target candidates.
Machine Learning & AI — Target Prediction Based on Learning Models
ML models learn from known ligand–protein data, then predict DHEA's ability to bind to new proteins based on chemical features (fingerprint, descriptor) or 3D structure.
SwissTargetPrediction
FreeInput DHEA SMILES, predict protein targets by comparing fingerprint with thousands of known ligand–target pairs. Outputs target list with probability and pathway.
SEA (Similarity Ensemble Approach)
FreeCompare DHEA with ligand sets for each target. Uses Tanimoto similarity + E-value statistics. Especially good for steroid molecules with known steroid analog targets.
TargetNet
FreeDeep learning model predicting targets from molecular fingerprint. Covers >600 human targets. Outputs binding probability for each protein.
DeepPurpose
FreeOpen-source Python framework with multiple DL architectures (Transformer, CNN, GNN) for DTI (Drug-Target Interaction). Can be fine-tuned with steroid data.
AlphaFold + docking
FreeAlphaFold2 predicts 3D structure of proteins without PDB, then use as receptor for docking. Especially useful when candidate proteins lack experimental structures.
DiffDock / RoseTTAFold-AA
FreeNext-gen AI models (diffusion-based) predict ligand–protein binding poses without predefined binding pocket. Promising for DHEA with steroid-binding proteins of unknown pocket.
💡 Start with SwissTargetPrediction and SEA as they're easiest to use (just need SMILES). Combine results from 2–3 tools, take the intersection as priority list.
Network Pharmacology & Omics — Network Analysis and Omics Data
Systems biology approach: find proteins in the protein interaction network (PPI network) located at key nodes related to DHEA's known biological effects.
Network pharmacology
STRING database
FreeBuild PPI network of known DHEA targets. Find hub proteins in the network. Hubs are often the most important targets.
Cytoscape
FreeVisualize and analyze protein networks. Calculate betweenness centrality, degree to identify hub proteins. BiNGO, ClueGO plugins for GO enrichment.
GEO / TCGA Omics
FreeTranscriptomics data analysis: find genes/proteins with expression changes after DHEA treatment. Changed expression + related pathway → candidate.
Enrichr / GSEA
FreePathway enrichment analysis from gene/protein lists. Find pathways affected by DHEA → narrow target list by pathway.
| Analysis Direction | Data Needed | Expected Result |
|---|---|---|
| Transcriptomics after DHEA | RNA-seq or published GEO dataset | List of up/down-regulated genes |
| PPI network of known targets | STRING + DHEA target list from ChEMBL | Hub proteins in network → priority targets |
| Similar binding site proteins | SiteMap / SiteAlign + PDB | Proteins with pockets similar to SHBG, AR |
| GWAS / eQTL | GTEx, UK Biobank | Genes genetically linked to DHEA levels |
💡 Thermal proteome profiling (TPP) — omics method similar to DARTS. If DARTS + MS is not possible, refer to published TPP data on ProteomicsDB or Thermal Proteome Atlas. Some steroids have been profiled — find proteins with similar thermal stability to DHEA.
3D Coordinate Optimization — Ligand Preparation for Docking
Before docking, ligand preparation generates realistic 3D conformations from 2D structures (SMILES/SDF). Three method categories—distance geometry, molecular mechanics, and quantum mechanics—use different math and return different geometries and potential energies. Ensure correct stereochemistry, protonation states, and partial charges.
Optimization tools
RDKit
FreeDistance geometry (ETKDG) plus MMFF94/UFF force-field minimization. Industry standard for Python pipelines, ML data prep, and small-molecule conformer generation before docking.
OpenBabel
FreeMechanics and distance geometry with MMFF94, UFF, GAFF, and Ghemical force fields. Swiss Army knife for bulk file conversion (110+ formats) combined with fast 3D cleanup.
Omega (OpenEye)
PaidKnowledge-based distance geometry with custom strain-energy rules. Commercial high-throughput conformer generation for large compound libraries.
Avogadro
FreeMolecular mechanics optimization (MMFF94, UFF, GAFF) with interactive 3D visualization. Best for manually inspecting and adjusting ligand structures.
xtb (Grimme Group)
FreeSemi-empirical quantum mechanics (GFN2-xTB / GFN-FF). Fast quantum-level structure refinement and charge calculation between mechanics and full QM.
Gaussian / ORCA
PaidPure quantum mechanics (DFT, HF, semi-empirical). Highest accuracy for final candidate refinement, precise partial charges, and deep electronic analysis—too slow for large libraries.
| Method Category | Underlying Approach | Best Used For |
|---|---|---|
| Distance Geometry | Mathematical distance-matrix smoothing (e.g., ETKDG). | Generating diverse 3D conformer libraries. |
| Molecular Mechanics | Classical physics: bonds as springs, atoms as charged spheres. | Rapidly cleaning and preparing large molecule libraries (MMFF94, UFF). |
| Quantum Mechanics | Approximations of the Schrödinger equation. | Refining final candidates; precise charge and geometry. |
💡 Different tools can yield different 3D coordinates and non-comparable energy scores (MMFF94 vs UFF). For flexible docking (Vina, Glide), initial geometry matters less; for rigid docking, conformer quality is critical. The golden rule: correct chemistry (R/S stereocenters, pH 7.4 protonation, atom types) matters more than minor coordinate differences.
Protein Preparation — Receptor Setup for Molecular Docking
Protein preparation impacts docking success more than ligand preparation. Key steps include water removal (conserved vs bulk), co-factor handling, hydrogen addition with correct protonation states, and grid box generation. Docking scoring functions rely on electrostatic and steric calculations—flawed protein input yields incorrect results.
Preparation tools
PDB2PQR / PROPKA
FreeExceptional pKa calculation and protonation state assignment at specific pH levels. Excellent for high-throughput protonation cleaning before automated docking via command-line, web server, or Python integration.
UCSF Chimera / ChimeraX
FreeIntuitive GUI for manually deleting chains, adding hydrogens, and stripping complex solvents. Best for visualizing pockets, fixing small file errors, and manually deleting unwanted artifact chains.
Protein Preparation Wizard
PaidIndustry gold standard (Schrödinger). Complete automated workflow for H-bond network optimization and structural flips. Enterprise-grade rigorous protein refinement where maximum prediction accuracy is required.
MGLTools / ADFRsuite
FreeSpecifically designed to convert standard PDB files into PDBQT format required by AutoDock Vina. Direct setup for workflows centered around AutoDock-family engines.
LePro
FreeBlazing-fast, single-command automated tool that strips waters, adds hydrogens, and prepares structural grids. Best for building heavy, automated screening pipelines for thousands of proteins.
BioPython (PDB module)
FreeTotal code-level control over PDB file's text array to surgically isolate specific coordinates. Ideal for programmatically parsing and sanitizing raw PDB structures in custom AI/ML pipelines.
| Processing Step | Basic / Automated Approach | Advanced / Rigorous Approach |
|---|---|---|
| Water Removal | Blind Stripping: Deletes all water molecules (HETATM lines) indiscriminately. | Thermodynamic Analysis: Computes water stability (3D-RISM/WaterMap) or uses explicit hydrated docking. |
| Co-factor & Artifact Handling | Blanket Deletion: Wipes out all non-protein chains, risking deletion of vital metal catalytic ions. | Manual/Informed Curation: Identifies and retains native active-site metals (Zn²⁺, Mg²⁺) while stripping buffer agents. |
| Hydrogen Addition | Geometric Placement: Places hydrogens based on static bond lengths, assuming fixed orientations. | pKa & H-Bond Optimization: Predicts protonation shifts via neighbor environments and flips flexible groups (Asn/Gln/His). |
| Grid Box Generation | Manual/Blind Placement: Encloses whole protein or relies on human judgment with graphical sliders. | Ligand-Centered / AI-Driven: Automates box centers on co-crystallized ligands or uses geometric pocket-finding algorithms. |
💡 The most common mistake is blanket deletion of all water molecules. Structural waters form critical hydrogen-bond networks between protein and ligand. Use thermodynamic analysis tools (WaterMap, 3D-RISM) to identify stable waters. Always run pKa predictors (PROPKA) to assign correct hydrogen configurations at target pH.
Molecular Docking — Docking DHEA into Proteins
Docking simulates DHEA (ligand) binding into the binding pocket of each hypothetical target protein. Calculates affinity score (docking score / binding energy). Proteins with good scores → priority candidates for DARTS.
AutoDock Vina
FreeMost popular open-source docking software. Supports rigid and flexible docking. Suitable for docking DHEA into individual proteins from PDB.
Glide (Schrödinger)
PaidHigher precision than Vina, with Maestro graphical interface. Supports XP mode (extra precision). Commonly used in professional drug research.
GOLD (CCDC)
PaidCalculates flexibility of both ligand and protein well. Diverse scoring functions (GoldScore, ChemScore, ASP). Suitable when high accuracy is needed.
SwissDock
FreeOnline web server docking, no installation needed. Upload protein structure from PDB and DHEA structure. Results within hours.
GNINA
FreeDocking with integrated deep learning scoring function (CNN). Often gives better results than Vina for steroid molecules thanks to learning from real PDB structures.
rDock / rDock2
FreeSuitable for large-scale virtual screening across many proteins simultaneously. Faster than AutoDock Vina when screening hundreds of proteins.
Basic Docking Workflow for DHEA
Prepare DHEA structure (ligand)
Obtain DHEA SMILES or SDF structure from PubChem (CID: 5881) or RCSB PDB (Ligand ID: AND). Optimize 3D coordinates using RDKit or Open Babel, assign Gasteiger charges, and export as .pdbqt (for Vina) or .mol2.
Prepare protein (receptor)
Download PDB structure of candidate protein (e.g., SHBG, G6PD, HSP90, AKR1C...). Remove water, unnecessary co-factors. Add hydrogens. Prepare grid box around binding pocket.
Run docking and evaluate results
Run AutoDock Vina or GNINA. Filter results: ΔG ≤ −7.0 kcal/mol is a threshold worth attention for DHEA (MW ~288). View hydrogen bond, hydrophobic interactions with PLIP or Discovery Studio.
💡 With DHEA being a steroid, prioritize docking into proteins with hydrophobic binding pockets such as steroid-binding proteins, nuclear receptors, hydroxysteroid dehydrogenases, and steroid-metabolizing enzymes.