← Benchmarks | BioTarget

BioTarget Predictions BENCHMARK

Binding affinity, target identification, and mechanism of action — validated against CASF-2016 and ChEMBL benchmarks. Validated against published benchmarks.

91%
MoA Accuracy
ChEMBL validation
0.537
Pearson r
CASF-2016 (270 complexes)
10,065
Targets
5 kingdoms
1.90
MAE (pKi)
CASF-2016 binding affinity

Why these numbers matter more than they look

FluxTarget solves a fundamentally harder problem than every method it is compared against.

What every other method receives
✓ The experimentally resolved 3D co-crystal structure — the protein and ligand already bound together, solved by X-ray crystallography at atomic resolution
✓ The exact binding pose — where the ligand sits, how it is oriented, which contacts it makes
✓ DL methods additionally train on 19,000+ solved structures from PDBbind
They score a known answer. The hard part — finding the pose — is already done by the crystallographer.
What FluxTarget receives
1. A SMILES string — a flat, 2D text representation of the molecule (e.g. CC(=O)Oc1ccccc1C(=O)O)
2. A target name (e.g. “Thrombin”)
No 3D structure. No binding pose. No training examples. FluxTarget must derive everything from physics.

What FluxTarget does with those two inputs

⚖
Builds 3D ligand
Generates multiple MMFF-optimized 3D conformers from the SMILES string
⚙
Loads 3D protein
Retrieves the full PDB structure and identifies the binding site
⚡
Docks on GPU
Searches rotational and translational poses, evaluates 14+ physics modules from 3D coordinates
★
Predicts affinity
Returns pKi with full energy breakdown — every term from first-principles physics
FluxTarget performs a complete computational docking simulation — building molecules in 3D, searching for binding poses, and computing interaction energies on GPU — all from a text string. The methods it is compared against skip the hardest steps because their input already contains the answer.

Validated Capabilities

Binding Affinity Prediction — CASF-2016

  • Pearson r = 0.537 on 270 CASF-2016 complexes (Su et al. 2019)
  • MAE = 1.90 pKi, Bias = -0.03 pKi
  • 33% within 1.0 pKi, 62% within 2.0 pKi of experimental
  • Every prediction deterministic and fully auditable
  • Full GPU physics simulation: 3D conformer generation, pose search, 14+ interaction energy modules
  • 5,000+ predictions/sec (~300,000x faster than conventional docking)

Literature Context

These are not apples-to-apples comparisons. Every other method receives the experimentally solved 3D co-crystal structure — the protein with the ligand already bound, at atomic resolution. They score a known pose. FluxTarget builds the ligand in 3D from a SMILES string, searches for the binding pose computationally, and simulates binding physics on GPU — all from scratch. Most DL methods also train on 19,000+ solved structures.
Method Pearson r MAE (pKi) Required Input Training Data
DEEP LEARNING + 3D CO-CRYSTAL STRUCTURE + LARGE TRAINING SETS
graphDelta (GNN) 0.87 — 3D co-crystal (X-ray) PDBbind (19K+ complexes)
Kdeep (CNN) 0.85 — 3D co-crystal (X-ray) PDBbind (19K+ complexes)
GNINA (CNN) 0.82 ~1.0 3D co-crystal (X-ray) PDBbind (19K+ complexes)
CLASSICAL SCORING + 3D CO-CRYSTAL STRUCTURE
RF-Score v3 0.72 ~1.4 3D co-crystal (X-ray) PDBbind (4K+ complexes)
Glide SP 0.65 — 3D co-crystal (X-ray) Empirical (fitted to structures)
X-Score 0.61 — 3D co-crystal (X-ray) Empirical (fitted to structures)
AutoDock Vina 0.60–0.70 ~1.5–1.7 3D co-crystal (X-ray) Empirical (fitted to structures)
NO 3D STRUCTURE · NO TRAINING DATA
FluxTarget v6.6b 0.537 1.90 SMILES + target name None (physics-only)

All methods benchmarked on CASF-2016 scoring power (285 complexes). DL methods trained on PDBbind general/refined sets. FluxTarget MAE of 1.90 pKi is comparable to Vina (~1.5–1.7) despite using no 3D structural information.

Download per-complex results
270 CASF-2016 complexes · PDB ID, target, experimental pKd, predicted pKi, error
Download CSV

Target Identification

  • AUC = 0.980 on target identification benchmark
  • 10,065 targets across 5 biological kingdoms
  • Predicts likely protein targets for a given small molecule
  • Physics-only scoring — no ML

Mechanism of Action (MoA) Prediction

  • 91% accuracy on ChEMBL validation set
  • Predicts agonist/antagonist/inhibitor classification
  • Physics-only approach — no ML, unique to FluxMateria
  • Integrated with FluxTarget module

Planned

Selectivity Profiling

Off-target binding predictions and selectivity scoring. Status: Planned.

Try FluxTarget on Your Data

Predict binding affinities across 10,000+ targets with full interpretability. CASF-2016 validated.

← Back to Module Join Research Preview