TF3P

TF3P generates 3D molecular fingerprints using a deep capsular network to encode three-dimensional force field information for ligand-based drug discovery.

Key Features:

• Deep Capsular Network Integration: Uses a deep capsular network to encode complex three-dimensional force field information of small molecules.
• Unsupervised Learning: Trains in an unsupervised manner and does not require labeled datasets specific to predictive tasks.
• 3D Structural Sensitivity: Captures subtle three-dimensional structural changes enabling recognition of compounds similar in 3D but not in 2D.
• Target Similarity Identification: Identifies targets with similar biological activity based on ligand similarity that may be missed by other 2D or 3D fingerprints.
• Model Compatibility: Produces descriptors compatible with statistical approaches such as the similarity ensemble approach and with machine learning models.

Scientific Applications:

• Ligand-based similarity searching: Enables 3D force field-based similarity searches to prioritize small molecules with comparable three-dimensional properties.
• Target identification by ligand similarity: Facilitates identification of biologically similar targets through ligand-based similarity comparisons.
• Integration into predictive modeling: Serves as input descriptors for statistical and machine learning models used in drug discovery and virtual screening.

Methodology:

Training and encoding are performed using an unsupervised deep capsular network to produce fingerprints that represent 3D force fields of small molecules.