DeepPINCS

DeepPINCS predicts compound-protein interactions (CPI) using a deep learning model that takes sequence information from compounds represented as SMILES strings and proteins represented as amino acid sequences to predict interaction strength or binary interaction outcomes.

Key Features:

• Sequence-based input encoding: Processes compounds as SMILES strings and proteins as amino acid sequences without requiring structural information.
• Compound and protein encoders: Implements dedicated encoders with useful features for both molecular and biological sequence data.
• Extended interaction modeling: Supports protein-protein interaction (PPI), chemical-chemical interaction (CCI), and single compound or protein modeling tasks.
• Nucleic acid compatibility: Accepts DNA and RNA sequences when represented as sequence data, extending applicability beyond proteins.

Scientific Applications:

• Drug discovery screening: Identifies novel compound-protein interactions to support design of new drugs for target proteins by screening candidate compounds.
• Interaction prediction: Predicts interaction strength or binary interaction outcomes between compounds and proteins using only sequence information.

Methodology:

The model uses deep learning with compound and protein encoders that process discrete symbolic sequence data; compounds are encoded from SMILES strings and proteins from amino acid sequences, with the outcome variable indicating interaction strength or presence of interaction.