Modeling drug combination effects via latent tensor reconstruction

Abstract

Motivation: Combination therapies have emerged as a powerful treatment modality to overcome drug resistance and improve treatment efficacy. However, the number of possible drug combinations increases very rapidly with the number of individual drugs in consideration, which makes the comprehensive experimental screening infeasible in practice. Machine-learning models offer time- and cost-efficient means to aid this process by prioritizing the most effective drug combinations for further pre-clinical and clinical validation. However, the complexity of the underlying interaction patterns across multiple drug doses and in different cellular contexts poses challenges to the predictive modeling of drug combination effects.

Results: We introduce comboLTR, highly time-efficient method for learning complex, non-linear target functions for describing the responses of therapeutic agent combinations in various doses and cancer cell-contexts. The method is based on a polynomial regression via powerful latent tensor reconstruction. It uses a combination of recommender system-style features indexing the data tensor of response values in different contexts, and chemical and multi-omics features as inputs. We demonstrate that comboLTR outperforms state-of-the-art methods in terms of predictive performance and running time, and produces highly accurate results even in the challenging and practical inference scenario where full dose-response matrices are predicted for completely new drug combinations with no available combination and monotherapy response measurements in any training cell line.

Availability and implementation: comboLTR code is available at https://github.com/aalto-ics-kepaco/ComboLTR.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

Illustration of the drug combination response tensor and its feature representation. (a) Drug combination responses form a fifth-order tensor indexed by drugs, their concentrations and the cell lines. (b) The drug combination response tensor can be flattened into a tensor index featurematrix via one-hot encoding and accompanied by chemical and biological information

Fig. 2.

Illustration of different drug combination response prediction scenarios. (a) Filling in the gaps in partially measured dose–response matrices (S1); predicting dose–response matrices of new drug combinations (b) with monotherapy responses (S2) and (c) without monotherapy responses (S3)

Fig. 3.

Predictive performance of comboLTR, comboFM and RF in three drug combination response prediction scenarios. Scatter plots between the predicted and measured dose–dependent drug combination effects in the form of %-growth of cancer cell lines. The predictions were made under three scenarios of (a) filling in the gaps in partially measured dose–response matrices, inferring dose–response matrices of completely new drug combinations with (b) and without (c) monotherapy responses available. Root mean squared error, Pearson correlation and Spearman correlation are reported as averages ± SDs over five CV folds. Diagonal line and linear fit are also displayed in each scatter plot. Note, different x- and y-axes ranges in the plots that are consistent across the panels

Fig. 4.

Predictive performance of comboLTR, comboFM and RF across tissue types and drug classes in three drug combination response prediction scenarios. Violin plots were used to characterize Pearson correlations of predicted and measured drug combination responses across tissue types (a–c) and drug classes (d–f). Note that the order of tissue types and drug classes in the legends corresponds to their order in the violin plots