# mdxplain Orchestration This table provides an overview of the methods, algorithms, and backends used in mdxplain, organized by category.

Category	Method/Algorithm	Backend	Explanation
Trajectory	Trajectory handling	mdtraj + mdxplain	mdxplain: In memmap-mode uses DaskTrajectory - Dask/Zarr manages data streaming and storage. mdtraj: Reads topology and frames, performs geometric operations (superpose, smooth, align, distances) on each loaded chunk. In normal mode uses mdtraj trajectories.
Nomenclature/Labeling	Consensus labeling (GPCR/CGN/KLIFS)	mdciao	Consensus labels come from mdciao; mdxplain maps them to features/selectors and integrates them into its internal metadata system.
Nomenclature/Labeling	Standard labeling and metadata generation	mdxplain	mdxplain implements own metadata / AA labeling using topology from mdtraj.
Feat. Sel. DSL	Feature Selector/Internal DSL	mdxplain + mdtraj	Whole Selection DSL by mdxplain. For atom-selection from topology mdtraj DSL is used.
Data Handling	Archiving and large data handling	Pickle	Archiving and saving implemented by mdxplain using Pickle.
Data Handling	Memmap arrays and low-level arithmetic	NumPy	Intermediate data is saved as NumPy memmap arrays. If not stated differently, NumPy also used for basic arithmetics like mean, median, var, etc.
Geometric Features	Coordinates	mdxplain	Whole feature type by mdxplain, trajectory object provides coordinates & topology.
	Distances	mdtraj	mdxplain orchestrates uses mdtraj.compute_contacts as computation backend.
	Contacts	mdxplain	Whole feature type by mdxplain, uses distance outputs (mdtraj kernel).
	Torsions	mdtraj	mdxplain orchestrates and uses mdtraj.compute_phi/psi/omega/chi as backend.
	DSSP	mdtraj	mdxplain orchestrates and uses mdtraj.dssp as backend.
	SASA	mdtraj	mdxplain orchestrates and uses mdtraj.shrake_rupley as backend.
	Reduction Logic	mdxplain + SciPy	mdxplain implements feature-reduction and uses NumPy/SciPy for metrics where possible, extending them when needed.
	RMSD	mdxplain	The trajectory object provides topology/coordinates; mdxplain handles computation plus data/metadata, including window and reference metrics (frame-to-frame or to-reference) and variants like RMSD with mean or median for flexible systems.
	RMSF	mdxplain + Numba-JIT	Trajectory supplies topology/coords; mdxplain handles computation and metadata, including window/reference metrics and RMSF variants (mean/median). It also performs residue-level aggregation (mean, median, RMS, RMdS) with Numba-JIT acceleration.
	MAD	mdxplain	mdxplain offers MAD values in frame-based and atom- / residue-based manner. This is implemented by mdxplain directly inside the RMSD and RMSF calculators.
Dim Reduction	PCA/IncrementalPCA	scikit-learn	PCA and incremental PCA wrap scikit-learn; mdxplain orchestrates.
	Kernel PCA	scikit-learn + scipy + mdxplain	Standard KPCA uses scikit-learn; iterative KPCA and metadata/data handling are implemented by mdxplain. Nystroem KPCA combines scikit-learn's RBF Nystroem with IncrementalPCA, with mdxplain adding component/epsilon approximations.
	Diffusion Maps	mdxplain	Full algorithm incl. iterative and nystroem mode implemented by mdxplain.
Clustering	DBSCAN	scikit-learn	mdxplain orchestrates DBSCAN: standard mode uses scikit-learn's DBSCAN, precomputed mode uses scikit-learn's NearestNeighbors, and kNN-mode combines scikit-learn's DBSCAN on a subsample with scikit-learn's kNN for assignment.
	HDBSCAN	hdbscan + scikit-learn	mdxplain orchestrates HDBSCAN workflows: standard mode uses the hdbscan implementation, approximate-prediction mode uses random sampling plus hdbscan's own prediction, and kNN-mode combines hdbscan on a subsample with scikit-learn's kNN for assignment.
	DPA	dpa + scikit-learn	mdxplain orchestrates DPA workflows: standard mode uses the dpa library's DPA implementation, and kNN-mode runs DPA on a subsample and assigns remaining points via scikit-learn's kNN.
Feature Importance	Decision Tree	scikit-learn	mdxplain orchestrates decision-tree workflows: it uses scikit-learn's DecisionTreeClassifier, adds stratified sampling for large datasets, and provides comparison modes (one-vs-rest, multiclass, binary, pairwise).
Feature Statistics	Feature-type analysis	mdxplain	mdxplain implements feature-based analysis and uses NumPy/SciPy for metrics where possible, extending them when needed.
Plots	DecisionTree-Plotter	mdxplain + matplotlib	Layout, styling, export by mdxplain. Uses tree-structure of scikit-learn model. Plot-library is matplotlib.
	DensityPlotter	mdxplain + matplotlib + scipy	mdxplain implementation using matplotlib. DensityPlotter uses scipy KDEs to display smoothed histrograms.
	ViolinPlotter	mdxplain + matplotlib	mdxplain implementation using matplotlib.
	TimeSeries-Plotter	mdxplain + matplotlib	mdxplain implementation using matplotlib. Time-series can be smoothed using a savgol filter from scipy.
	Landscape-Plotter	mdxplain + matplotlib + scipy	mdxplain implementation using matplotlib. Uses NumPy for histograms and SciPy for optional KDE smoothing.
	Cluster-Membership-Plotter	mdxplain + matplotlib	mdxplain implementation using matplotlib.
3D Viz	StructureViz Feature Service	mdxplain + mdtraj + PyMol	mdxplain implements representative-finding (centroid or decision-tree–based), uses mdtraj to generate PDBs. mdxplain embed feature-importance values as B-factors, and provides PyMOL script generation for optional visualization. PyMOL can be used optional to visualize this script.
3D Viz	NGLView	mdxplain + mdtraj + nglview	mdxplain implements representative-finding (centroid or decision-tree–based), uses mdtraj to generate PDBs. mdxplain embed feature-importance values as B-factors, and provides an extended nglview widget with checkboxes, selections, and legends for PyMOL-like 3D comparison in Jupyter.