Predicting drug response of tumors from integrated genomic profiles by deep neural networks (1805.07702v1)

Abstract: The study of high-throughput genomic profiles from a pharmacogenomics viewpoint has provided unprecedented insights into the oncogenic features modulating drug response. A recent screening of ~1,000 cancer cell lines to a collection of anti-cancer drugs illuminated the link between genotypes and vulnerability. However, due to essential differences between cell lines and tumors, the translation into predicting drug response in tumors remains challenging. Here we proposed a DNN model to predict drug response based on mutation and expression profiles of a cancer cell or a tumor. The model contains a mutation and an expression encoders pre-trained using a large pan-cancer dataset to abstract core representations of high-dimension data, followed by a drug response predictor network. Given a pair of mutation and expression profiles, the model predicts IC50 values of 265 drugs. We trained and tested the model on a dataset of 622 cancer cell lines and achieved an overall prediction performance of mean squared error at 1.96 (log-scale IC50 values). The performance was superior in prediction error or stability than two classical methods and four analog DNNs of our model. We then applied the model to predict drug response of 9,059 tumors of 33 cancer types. The model predicted both known, including EGFR inhibitors in non-small cell lung cancer and tamoxifen in ER+ breast cancer, and novel drug targets. The comprehensive analysis further revealed the molecular mechanisms underlying the resistance to a chemotherapeutic drug docetaxel in a pan-cancer setting and the anti-cancer potential of a novel agent, CX-5461, in treating gliomas and hematopoietic malignancies. Overall, our model and findings improve the prediction of drug response and the identification of novel therapeutic options.

• The paper presents a deep neural network framework that integrates mutation and expression data to predict log-scale IC50 values for over 265 anti-cancer drugs.
• It demonstrates superior performance with a mean squared error of 1.96, effectively capturing established and novel drug-gene interactions.
• The research highlights the clinical potential of genomics-driven models to personalize cancer treatments and advance precision oncology.

Insights into Predicting Tumor Drug Response Using Deep Neural Networks

The paper titled "Predicting drug response of tumors from integrated genomic profiles by deep neural networks" presents a compelling approach to predicting tumor responses to anti-cancer drugs using deep neural networks (DNNs). This research leverages high-dimensional genomic data to enhance the prediction of drug efficacy in tumors, a notable challenge in precision oncology.

Methodology and Model Architecture

The paper outlines a DNN framework comprising three subnetworks: a mutation encoder, an expression encoder, and a drug response predictor network. Both the mutation and expression encoders are pre-trained using The Cancer Genome Atlas (TCGA) dataset to extract core features from mutation and expression data. The final drug response predictor integrates these features to predict log-scale IC50 values for 265 drugs. This model was rigorously trained using a dataset of 622 Cancer Cell Line Encyclopedia (CCLE) cell lines and evaluated on 9,059 tumors across 33 cancer types.

The pre-trained encoders are used to effectively abstract high-order features from mutation and expression data, reducing their dimensions while retaining critical discriminative information. This approach addresses the complexity of genomic data and improves the model's learning efficiency. The paper further undertakes comparisons with traditional methods like linear regression, support vector machines (SVM), and other DNN architectures, demonstrating superior performance with an mean squared error (MSE) of 1.96 in predicting IC50 values.

Results and Applications

One salient result from the model is its ability to capture known drug-gene interactions, such as EGFR inhibitors in non-small cell lung cancer and tamoxifen in ER-positive breast cancer. Additionally, it suggests novel therapeutic targets like the efficacy of vinorelbine in TTN-mutated tumors. This predictive capability was further explored to reveal drug resistance mechanisms, for instance, resistance to docetaxel and the identification of potential new drug applications for agents such as CX-5461 in gliomas and hematopoietic malignancies.

The model's validation across diverse cancer types elucidates its practical application in identifying effective drug targets and resistance mechanisms in a tumor-specific manner. This underscores its potential in facilitating personalized treatment regimens based on individual tumor genomics.

Implications and Future Prospects

The integration of pharmacogenomic data using DNNs not only improves predictive accuracy but opens avenues for identifying uncharted associations between genomic alterations and drug responses. The robustness of this model suggests its utility in enhancing clinical decision-making processes by offering a genomics-driven perspective on therapy selection.

From a theoretical standpoint, this research advances our understanding of the interplay between genomic profiles and therapeutic efficacy, reinforcing the value of sophisticated computational models in unraveling complex biological interactions. Future work could further enhance this model by incorporating other omics data, refining genomic feature extraction techniques, and addressing interpretability issues inherent in DNNs. As more comprehensive datasets become available, the model's predictive capacity and clinical applicability are bound to improve, inching closer to the long-sought goal of precision oncology.

In conclusion, this paper marks a significant step in utilizing deep learning frameworks to bridge the gap between genomic data and clinical oncology, providing a scalable and effective tool for the prediction of tumor drug response. As the field progresses, such models are poised to become integral components of personalized cancer treatment strategies.