comboLTR

comboLTR predicts anti-cancer effects of drug combinations by modeling nonlinear dose–response interactions across cancer cell lines using polynomial regression with latent tensor reconstruction.

Key Features:

• Polynomial Regression via Latent Tensor Reconstruction: Implements polynomial regression through latent tensor reconstruction to model nonlinear target functions across dose and context dimensions.
• Recommender System-Style Features: Indexes response data as tensors across diverse contexts using recommender-system-style features to capture context-specific interaction patterns.
• Integration of Chemical and Multi-Omics Features: Incorporates chemical properties and multi-omics data as input features to inform predictions.
• Efficiency and Predictive Performance: Predicts full dose–response matrices for novel drug combinations without requiring prior monotherapy or combination response data and reports improved predictive accuracy and computational efficiency relative to state-of-the-art methods.

Scientific Applications:

• Prioritization of Drug Combinations: Prioritizes candidate drug combinations for pre-clinical and clinical validation by predicting combination efficacy across doses and cell lines.
• Overcoming Drug Resistance: Models combination therapies to support identification of regimens that may overcome drug resistance in cancer.

Methodology:

Constructs a data tensor of response values across doses and cell lines, incorporates chemical properties and multi-omics features, indexes tensors with recommender-system-style features, and applies polynomial regression via latent tensor reconstruction to predict dose–response matrices for new drug combinations.