QuBiLS-MIDAS

QuBiLS-MIDAS computes alignment-free three-dimensional molecular descriptors (3D-MDs) that encode atom-pair and higher-order geometrical relationships for molecular characterization and predictive modeling.

Key Features:

• Advanced descriptor calculation: Computes chiral, fuzzy, weighted, and truncated geometrical 3D-MDs using tensor algebra in two-linear (bilinear), three-lineal, and four-linear (multi-linear or N-linear) forms.
• Atomic weighting: Incorporates atomic weightings derived from indices such as the vertex-degree invariant (e.g., Alikhanidi index) to modulate atom contributions.
• Chirality encoding: Considers central chirality during molecular encoding to retain stereochemical information.
• Multiatomic structural codification: Employs clustering methods and statistical functions to encode structural relationships among more than two atoms.
• Fuzzy spherical truncation: Uses fuzzy membership functions to enable spherical truncation of inter-atomic relations.
• Aggregation operators: Applies weighted and fuzzy aggregation operators to compute global 3D-MDs from local relations.
• Modular descriptor computation: Provides a module to compute QuBiLS-MIDAS 3D-MDs from descriptor headings.
• Predefined descriptor set: Supplies a set of predefined 3D-MDs with high information content and low redundancy for over 20,469 compounds, showing superior performance in variability studies compared to Dragon (v5.5) and PaDEL 0D-to-3D MDs and providing orthogonal chemical information.

Scientific Applications:

• Drug discovery: Generates detailed 3D descriptors for molecular characterization and predictive modeling in drug discovery.
• Materials science: Provides 3D molecular descriptors for characterization and modeling of materials-related molecules.
• Toxicology: Produces descriptors suitable for toxicology modeling and hazard assessment.
• Predictive modeling and benchmarking: Supplies orthogonal, low-redundancy descriptors for predictive modeling and variability benchmarking against Dragon (v5.5) and PaDEL 0D-to-3D MDs.

Methodology:

Computes atom-pair and alignment-free 3D-MDs using various distance metrics beyond Euclidean distance, tensor-algebra encodings (bilinear, three-lineal, four-linear/N-linear), atomic weightings (vertex-degree invariant/Alikhanidi index), central chirality consideration, clustering methods and statistical functions for multi-atom relations, fuzzy membership functions for spherical truncation, and weighted/fuzzy aggregation operators; includes generation of predefined descriptors for >20,469 compounds and variability comparisons to Dragon (v5.5) and PaDEL 0D-to-3D MDs.