Uncovering expression signatures of synergistic drug response using an ensemble of explainable AI models

Abstract

Complex machine learning models are poised to revolutionize the treatment of diseases like acute myeloid leukemia (AML) by helping physicians choose optimal combinations of anti-cancer drugs based on molecular features. While accurate predictions are important, it is equally important to be able to learn about the underlying molecular basis of anti-cancer drug synergy. Explainable AI (XAI) offers a promising new route for data-driven cancer pharmacology, combining highly accurate models with interpretable insights into model decisions. Due to the highly correlated, high-dimensional nature of cancer transcriptomic data, however, we find that existing XAI approaches are suboptimal when applied naively to large transcriptomic datasets. We show how a novel approach based on model ensembling helps to increase the quality of explanations. We then use our method to demonstrate that a hematopoietic differentiation signature underlies synergy for a variety of anti-AML drug combinations.

A deep profile of gene expression across 18 human cancers

“…Moreover, the application of Explainable AI (XAI) in life sciences [21][22][23] , although widespread, often grapples with complex, multidimensional data. In this context, model ensembles offer a significant advantage, improving the quality and reliability of feature attribution 24 , thereby aligning with the growing emphasis on transparency and comprehension in AI models used for biological data analysis.…”

Machine learning classifier approaches for predicting response to RTK-type-III inhibitors demonstrate high accuracy using transcriptomic signatures and ex vivo data

“…However, regarding classification algorithms (CA) for predicting response to therapy, a scalable ML approach has not been efficiently identified. Recently, investigators have reported on support vector machines ( Castillo et al , 2019 ; Gal et al , 2019 ; Lee et al , 2018 ), k-nearest neighbor ( Castillo et al , 2019 ; Gal et al , 2019 ; Lee et al , 2018 ), RF ( Castillo et al , 2019 ; Gal et al , 2019 ), artificial neural networks ( Janizek et al , 2021 ) and gradient boosting ( Janizek et al , 2021 ) for conducting these types of analysis.…”

“…Many feature extraction algorithms, or rather feature selection techniques, for gene expression data have been proposed, from the traditional use of principal component analysis (PCA; Gal et al , 2019 ) to customized techniques ( Castillo et al , 2019 ; Lee et al , 2018 ). Some of these works include the EXPRESS ( Janizek et al , 2021 ) algorithm that uses ensembles of XGBoost models and the Shapley Additive Explanation (SHAP) package to calculate a global feature importance ranking. Yap et al (2021) used SHAP to explain a convolutional neural network to classify samples on 47 different tissues based on RNA-seq count data and compared the genes identified by SHAP with differential expression analysis.…”