Toward Explainable Anticancer Compound Sensitivity Prediction via Multimodal Attention-Based Convolutional Encoders

Manica, Matteo; Oskooei, Ali; Born, Jannis; Subramanian, Vigneshwari; Sáez-Rodríguez, Julio; Martínez, María Rodríguez

doi:10.1021/acs.molpharmaceut.9b00520

Cited by 108 publications

(132 citation statements)

References 74 publications

(163 reference statements)

Supporting

Mentioning

123

Contrasting

Order By: Relevance

“…First, interpretability of deep learning models is currently an open research topic of high interest. Some approaches to help interpret deep learning models are beginning to appear, such as the 'attention approach', which has been applied to biology (Manica et al, 2019). Second, microbiome data are characterized by high dimensionality (in terms of hundreds or thousands of different microbes) and few samples (usually up to hundreds).…”

Section: Discussionmentioning

confidence: 99%

“…However, these are difficult to interpret (beyond 'the lowest' and 'the best') and in this case would be difficult to compare because we apply different normalization approaches, confounding these metrics because they are scale-dependent. As such, we compute additional scale-independent metrics, such as Pearson correlation, that appear also in other bio-autoencoder systems (Manica et al, 2019). An additional metric selected was Bray-Curtis dissimilarity, described above in Loss Functions section.…”

Section: Evaluation Metricsmentioning

confidence: 99%

See 1 more Smart Citation

Predicting microbiomes through a deep latent space

García-Jiménez

Muñoz²,

Cabello

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

Section: Discussionmentioning

confidence: 99%

Section: Evaluation Metricsmentioning

confidence: 99%

Predicting microbiomes through a deep latent space

García-Jiménez

Muñoz²,

Cabello

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…More recently, deep neural networks with multiple hidden layers such as CDRscan [33], 34 tCNNS [34], and MCA [35] have been proposed that achieve R 2 higher than 0.8 (R 2 = 35 0.84, 0.83, and 0.86, respectively). However, most of the cancer cell line features used in 36 previous studies were based on gene expression profiles and did not explicitly consider 37 associations between drugs and the structural location of mutations (Table 1).…”

mentioning

confidence: 99%

“…10. 16 Manica et al [35] CNN + RNN (RF, SVM) EXP, CNV, PPI 5-fold CV R 2 = 0.86 2019.11.04 Oskooei et al [24] Network (RF, LR) EXP, PPI 30-fold CV r = ∼0.9 Fig 1. The framework of using the QSMART model with neural networks to predict protein kinase inhibitor response in cancer cell lines. Four main components of this framework: (1) drug features, cancer cell line features, and drug responses, (2) statistics tests for interaction terms, (3) a feature selection method for identifying highly informative features, and (4) a machine learning method for predicting drug response.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Quantitative Structure-Mutation-Activity Relationship Tests (QSMART) Model for Protein Kinase Inhibitor Response Prediction

Huang

Yeung

Wang

et al. 2019

Preprint

View full text Add to dashboard Cite

Predicting drug sensitivity profiles from genotypes is a major challenge in personalized medicine. Machine learning and deep neural network methods have shown promise in addressing this challenge, but the "black-box" nature of these methods precludes a mechanistic understanding of how and which genomic and proteomic features contribute to the observed drug sensitivity profiles. Here we provide a combination of statistical and neural network framework that not only estimates drug IC 50 in cancer cell lines with high accuracy (R 2 = 0.861 and RMSE = 0.818) but also identifies features contributing to the accuracy, thereby enhancing explainability. Our framework, termed QSMART, uses a multi-component approach that includes (1) collecting drug fingerprints, cancer cell line's multi-omics features, and drug responses, (2) testing the statistical significance of interaction terms, (3) selecting features by Lasso with Bayesian information criterion, and (4) using neural networks to predict drug response. We evaluate the contribution of each of these components and use a case study to explain the biological relevance of several selected features to protein kinase inhibitor response in non-small cell lung cancer cells. Specifically, we illustrate how interaction terms that capture associations between drugs and mutant kinases quantitatively contribute to the response of two EGFR inhibitors (afatinib and lapatinib) in non-small cell lung cancer cells. Although we have tested QSMART on protein kinase inhibitors, it can be extended across the proteome to investigate the complex relationships connecting genotypes and drug sensitivity profiles. Introduction 1 Protein kinases are a class of signaling proteins, greatly valued as therapeutic targets for 2 their key roles in human diseases, such as cancer [1]. For decades, chemotherapy has 3 served as part of a standard set of cancer treatments; however, the resistance of cancer 4 cells to chemotherapy is still a major clinical challenge [2]. Mutations in protein kinase 5 are known to play important roles not only in drug resistance [3] but also in drug 6 sensitivity [4]. Depending on the structural location, mutations can have varying 7 impacts on drug sensitivity. For example, non-small cell lung cancer (NSCLC) cells 8 December 28, 2019 1/25 harboring either the EGFR T790M or L858R mutation respectively leads to resistance 9 or hypersensitivity to the cancer drug gefitinib [5, 6], while those with EGFR 10 T790M/L858R double mutant are only resistant to gefitinib [7]. As mutations impact 11 the efficacy of different cancer drugs, there is a need to incorporate structural 12 knowledge in drug response prediction methods. 13 To facilitate the understanding of the molecular mechanisms that cause drug 14 sensitivity and drug resistance in cancer cells, the Genomics of Drug Sensitivity in 15 Cancer (GDSC) Project [8] recently screened the drug responses of 266 anti-cancer 16 drugs against ∼1,000 human cancer cell lines and provided the largest publicly available 17 drug response datas...

show abstract

Enabling Single‐Cell Drug Response Annotations from Bulk RNA‐Seq Using SCAD

Zheng

Chen²,

Chen

et al. 2023

Advanced Science

View full text Add to dashboard Cite

The single-cell RNA sequencing (scRNA-seq) quantifies the gene expression of individual cells, while the bulk RNA sequencing (bulk RNA-seq) characterizes the mixed transcriptome of cells. The inference of drug sensitivities for individual cells can provide new insights to understand the mechanism of anti-cancer response heterogeneity and drug resistance at the cellular resolution. However, pharmacogenomic information related to their corresponding scRNA-Seq is often limited. Therefore, a transfer learning model is proposed to infer the drug sensitivities at single-cell level. This framework learns bulk transcriptome profiles and pharmacogenomics information from population cell lines in a large public dataset and transfers the knowledge to infer drug efficacy of individual cells. The results suggest that it is suitable to learn knowledge from pre-clinical cell lines to infer pre-existing cell subpopulations with different drug sensitivities prior to drug exposure. In addition, the model offers a new perspective on drug combinations. It is observed that drug-resistant subpopulation can be sensitive to other drugs (e.g., a subset of JHU006 is Vorinostat-resistant while Gefitinib-sensitive); such finding corroborates the previously reported drug combination (Gefitinib + Vorinostat) strategy in several cancer types. The identified drug sensitivity biomarkers reveal insights into the tumor heterogeneity and treatment at cellular resolution.

show abstract

Toward Explainable Anticancer Compound Sensitivity Prediction via Multimodal Attention-Based Convolutional Encoders

Abstract: Equal contribution

Cited by 108 publications

References 74 publications

Predicting microbiomes through a deep latent space

Predicting microbiomes through a deep latent space

Quantitative Structure-Mutation-Activity Relationship Tests (QSMART) Model for Protein Kinase Inhibitor Response Prediction

Enabling Single‐Cell Drug Response Annotations from Bulk RNA‐Seq Using SCAD

Contact Info

Product

Resources

About