Deep learning of pharmacogenomics resources: moving towards precision oncology.

The recent accumulation of cancer genomic data provides an opportunity to understand how a tumor's genomic characteristics can affect its responses to drugs. This field, called pharmacogenomics, is a key area in the development of precision oncology. Deep learning (DL) methodology has emerged as a powerful technique to characterize and learn from rapidly accumulating pharmacogenomics data. We introduce the fundamentals and typical model architectures of DL. We review the use of DL in classification of cancers and cancer subtypes (diagnosis and treatment stratification of patients), prediction of drug response and drug synergy for individual tumors (treatment prioritization for a patient), drug repositioning and discovery and the study of mechanism/mode of action of treatments. For each topic, we summarize current genomics and pharmacogenomics data resources such as pan-cancer genomics data for cancer cell lines (CCLs) and tumors, and systematic pharmacologic screens of CCLs. By revisiting the published literature, including our in-house analyses, we demonstrate the unprecedented capability of DL enabled by rapid accumulation of data resources to decipher complex drug response patterns, thus potentially improving cancer medicine. Overall, this review provides an in-depth summary of state-of-the-art DL methods and up-to-date pharmacogenomics resources and future opportunities and challenges to realize the goal of precision oncology.

Lineage of origin in rhabdomyosarcoma informs pharmacological response.

Lineage or cell of origin of cancers is often unknown and thus is not a consideration in therapeutic approaches. Alveolar rhabdomyosarcoma (aRMS) is an aggressive childhood cancer for which the cell of origin remains debated. We used conditional genetic mouse models of aRMS to activate the pathognomonic Pax3:Foxo1 fusion oncogene and inactivate p53 in several stages of prenatal and postnatal muscle development. We reveal that lineage of origin significantly influences tumor histomorphology and sensitivity to targeted therapeutics. Furthermore, we uncovered differential transcriptional regulation of the Pax3:Foxo1 locus by tumor lineage of origin, which led us to identify the histone deacetylase inhibitor entinostat as a pharmacological agent for the potential conversion of Pax3:Foxo1-positive aRMS to a state akin to fusion-negative RMS through direct transcriptional suppression of Pax3:Foxo1.

CeL-ID: cell line identification using RNA-seq data.

BACKGROUND: Cell lines form the cornerstone of cell-based experimentation studies into understanding the underlying mechanisms of normal and disease biology including cancer. However, it is commonly acknowledged that contamination of cell lines is a prevalent problem affecting biomedical science and available methods for cell line authentication suffer from limited access as well as being too daunting and time-consuming for many researchers. Therefore, a new and cost effective approach for authentication and quality control of cell lines is needed. RESULTS: We have developed a new RNA-seq based approach named CeL-ID for cell line authentication. CeL-ID uses RNA-seq data to identify variants and compare with variant profiles of other cell lines. RNA-seq data for 934 CCLE cell lines downloaded from NCI GDC were used to generate cell line specific variant profiles and pair-wise correlations were calculated using frequencies and depth of coverage values of all the variants. Comparative analysis of variant profiles revealed that variant profiles differ significantly from cell line to cell line whereas identical, synonymous and derivative cell lines share high variant identity and are highly correlated (ρ > 0.9). Our benchmarking studies revealed that CeL-ID method can identify a cell line with high accuracy and can be a valuable tool of cell line authentication in biomedical science. Finally, CeL-ID estimates the possible cross contamination using linear mixture model if no perfect match was detected. CONCLUSIONS: In this study, we show the utility of an RNA-seq based approach for cell line authentication. Our comparative analysis of variant profiles derived from RNA-seq data revealed that variant profiles of each cell line are distinct and overall share low variant identity with other cell lines whereas identical or synonymous cell lines show significantly high variant identity and hence variant profiles can be used as a discriminatory/identifying feature in cell authentication model.

An ancestry informative marker panel design for individual ancestry estimation of Hispanic population using whole exome sequencing data.

BACKGROUND: Europeans and American Indians were major genetic ancestry of Hispanics in the U.S. These ancestral groups have markedly different incidence rates and outcomes in many types of cancers. Therefore, the genetic admixture may cause biased genetic association study with cancer susceptibility variants specifically in Hispanics. For example, the incidence rate of liver cancer has been shown with substantial disparity between Hispanic, Asian and non-Hispanic white populations. Currently, ancestry informative marker (AIM) panels have been widely utilized with up to a few hundred ancestry-informative single nucleotide polymorphisms (SNPs) to infer ancestry admixture. Notably, current available AIMs are predominantly located in intron and intergenic regions, while the whole exome sequencing (WES) protocols commonly used in translational research and clinical practice do not cover these markers. Thus, it remains challenging to accurately determine a patient's admixture proportion without additional DNA testing. RESULTS: In this study we designed an unique AIM panel that infers 3-way genetic admixture from three distinct and selective continental populations (African (AFR), European (EUR), and East Asian (EAS)) within evolutionarily conserved exonic regions. Initially, about 1 million exonic SNPs from selective three populations in the 1000 Genomes Project were trimmed by their linkage disequilibrium (LD), restricted to biallelic variants, and finally we optimized to an AIM panel with 250 SNP markers, or the UT-AIM250 panel, using their ancestral informativeness statistics. Comparing to published AIM panels, UT-AIM250 performed better accuracy when we tested with three ancestral populations (accuracy: 0.995 ± 0.012 for AFR, 0.997 ± 0.007 for EUR, and 0.994 ± 0.012 for EAS). We further demonstrated the performance of the UT-AIM250 panel to admixed American (AMR) samples of the 1000 Genomes Project and obtained similar results (AFR, 0.085 ± 0.098; EUR, 0.665 ± 0.182; and EAS, 0.250 ± 0.205) to previously published AIM panels (Phillips-AIM34: AFR, 0.096 ± 0.127, EUR, 0.575 ± 0.290, and EAS, 0.330 ± 0.315; Wei-AIM278: AFR, 0.070 ± 0.096, EUR, 0.537 ± 0.267, and EAS, 0.393 ± 0.300). Subsequently, we applied the UT-AIM250 panel to a clinical dataset of 26 self-reported Hispanic patients in South Texas with hepatocellular carcinoma (HCC). We estimated the admixture proportions using WES data of adjacent non-cancer liver tissues (AFR, 0.065 ± 0.043; EUR, 0.594 ± 0.150; and EAS, 0.341 ± 0.160). Similar admixture proportions were identified from corresponding tumor tissues. In addition, we estimated admixture proportions of The Cancer Genome Atlas (TCGA) collection of hepatocellular carcinoma (TCGA-LIHC) samples (376 patients) using the UT-AIM250 panel. The panel obtained consistent admixture proportions from tumor and matched normal tissues, identified 3 possible incorrectly reported race/ethnicity, and/or provided race/ethnicity determination if necessary. CONCLUSIONS: Here we demonstrated the feasibility of using evolutionarily conserved exonic regions to infer admixture proportions and provided a robust and reliable control for sample collection or patient stratification for genetic analysis. R implementation of UT-AIM250 is available at https://github.com/chenlabgccri/UT-AIM250.

Detection of high variability in gene expression from single-cell RNA-seq profiling.

BACKGROUND: The advancement of the next-generation sequencing technology enables mapping gene expression at the single-cell level, capable of tracking cell heterogeneity and determination of cell subpopulations using single-cell RNA sequencing (scRNA-seq). Unlike the objectives of conventional RNA-seq where differential expression analysis is the integral component, the most important goal of scRNA-seq is to identify highly variable genes across a population of cells, to account for the discrete nature of single-cell gene expression and uniqueness of sequencing library preparation protocol for single-cell sequencing. However, there is lack of generic expression variation model for different scRNA-seq data sets. Hence, the objective of this study is to develop a gene expression variation model (GEVM), utilizing the relationship between coefficient of variation (CV) and average expression level to address the over-dispersion of single-cell data, and its corresponding statistical significance to quantify the variably expressed genes (VEGs). RESULTS: We have built a simulation framework that generated scRNA-seq data with different number of cells, model parameters, and variation levels. We implemented our GEVM and demonstrated the robustness by using a set of simulated scRNA-seq data under different conditions. We evaluated the regression robustness using root-mean-square error (RMSE) and assessed the parameter estimation process by varying initial model parameters that deviated from homogeneous cell population. We also applied the GEVM on real scRNA-seq data to test the performance under distinct cases. CONCLUSIONS: In this paper, we proposed a gene expression variation model that can be used to determine significant variably expressed genes. Applying the model to the simulated single-cell data, we observed robust parameter estimation under different conditions with minimal root mean square errors. We also examined the model on two distinct scRNA-seq data sets using different single-cell protocols and determined the VEGs. Obtaining VEGs allowed us to observe possible subpopulations, providing further evidences of cell heterogeneity. With the GEVM, we can easily find out significant variably expressed genes in different scRNA-seq data sets.

Differential expression analysis of RNA sequencing data by incorporating non-exonic mapped reads.

BACKGROUND: RNA sequencing (RNA-seq) is a powerful tool for genome-wide expression profiling of biological samples with the advantage of high-throughput and high resolution. There are many existing algorithms nowadays for quantifying expression levels and detecting differential gene expression, but none of them takes the misaligned reads that are mapped to non-exonic regions into account. We developed a novel algorithm, XBSeq, where a statistical model was established based on the assumption that observed signals are the convolution of true expression signals and sequencing noises. The mapped reads in non-exonic regions are considered as sequencing noises, which follows a Poisson distribution. Given measureable observed and noise signals from RNA-seq data, true expression signals, assuming governed by the negative binomial distribution, can be delineated and thus the accurate detection of differential expressed genes. RESULTS: We implemented our novel XBSeq algorithm and evaluated it by using a set of simulated expression datasets under different conditions, using a combination of negative binomial and Poisson distributions with parameters derived from real RNA-seq data. We compared the performance of our method with other commonly used differential expression analysis algorithms. We also evaluated the changes in true and false positive rates with variations in biological replicates, differential fold changes, and expression levels in non-exonic regions. We also tested the algorithm on a set of real RNA-seq data where the common and different detection results from different algorithms were reported. CONCLUSIONS: In this paper, we proposed a novel XBSeq, a differential expression analysis algorithm for RNA-seq data that takes non-exonic mapped reads into consideration. When background noise is at baseline level, the performance of XBSeq and DESeq are mostly equivalent. However, our method surpasses DESeq and other algorithms with the increase of non-exonic mapped reads. Only in very low read count condition XBSeq had a slightly higher false discovery rate, which may be improved by adjusting the background noise effect in this situation. Taken together, by considering non-exonic mapped reads, XBSeq can provide accurate expression measurement and thus detect differential expressed genes even in noisy conditions.

Early gene expression in salivary gland after isoproterenol treatment.

Isoproterenol, a ß-adrenergic agonist, has been shown to induce salivary gland hyperplasia. However, the mechanism involved in this pharmacological phenomenon is not well understood. To gain a better understanding of the underlying changes, including genes, networks and pathways altered by isoproterenol, microarray-based gene expression analysis was conducted on rat parotid glands at 10, 30, and 60 min after isoproterenol injection. After isoproterenol treatment, the number of differentially expressed genes was increased in a time-dependent manner. Pathway analysis showed that cell hyperplasia, p38(MAPK), and IGF-1 were the most altered function, network and pathway, respectively. The balanced regulation of up- and down-expression of genes related to cell proliferation/survival may provide a better understanding of the mechanism of isoproterenol-induced parotid gland enlargement without tumor transformation.

A novel significance score for gene selection and ranking.

MOTIVATION: When identifying differentially expressed (DE) genes from high-throughput gene expression measurements, we would like to take both statistical significance (such as P-value) and biological relevance (such as fold change) into consideration. In gene set enrichment analysis (GSEA), a score that can combine fold change and P-value together is needed for better gene ranking. RESULTS: We defined a gene significance score π-value by combining expression fold change and statistical significance (P-value), and explored its statistical properties. When compared to various existing methods, π-value based approach is more robust in selecting DE genes, with the largest area under curve in its receiver operating characteristic curve. We applied π-value to GSEA and found it comparable to P-value and t-statistic based methods, with added protection against false discovery in certain situations. Finally, in a gene functional study of breast cancer profiles, we showed that using π-value helps elucidating otherwise overlooked important biological functions. AVAILABILITY: http://gccri.uthscsa.edu/Pi_Value_Supplementary.asp CONTACT: xy@ieee.org, cheny8@uthscsa.edu SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Integrative multi-omics characterization of hepatocellular carcinoma in Hispanic patients.

Background: The incidence and mortality rates of hepatocellular carcinoma (HCC) among Hispanics in the United States are much higher than those of non-Hispanic whites. We conducted comprehensive multi-omics analyses to understand molecular alterations in HCC among Hispanic patients. Methods: Paired tumor and adjacent non-tumor samples were collected from 31 Hispanic HCC in South Texas (STX-Hispanic) for genomic, transcriptomic, proteomic, and metabolomic profiling. Additionally, serum lipids were profiled in 40 Hispanic and non-Hispanic patients with or without clinically diagnosed HCC. Results: Exome sequencing revealed high mutation frequencies of AXIN2 and CTNNB1 in STX Hispanic HCCs, suggesting a predominant activation of the Wnt/ß-catenin pathway. The TERT promoter mutation frequency was also remarkably high in the Hispanic cohort. Cell cycles and liver functions were identified as positively- and negatively-enriched, respectively, with gene set enrichment analysis. Gene sets representing specific liver metabolic pathways were associated with dysregulation of corresponding metabolites. Negative enrichment of liver adipogenesis and lipid metabolism corroborated with a significant reduction in most lipids in the serum samples of HCC patients. Two HCC subtypes from our Hispanic cohort were identified and validated with the TCGA liver cancer cohort. The subtype with better overall survival showed higher activity of immune and angiogenesis signatures, and lower activity of liver function-related gene signatures. It also had higher levels of immune checkpoint and immune exhaustion markers. Conclusions: Our study revealed some specific molecular features of Hispanic HCC and potential biomarkers for therapeutic management of HCC and provides a unique resource for studying Hispanic HCC.

Integrative multi-omics characterization of hepatocellular carcinoma in hispanic patients.

BACKGROUND: The incidence and mortality rates of hepatocellular carcinoma (HCC) among Hispanic individuals in the United States are much higher than in non-Hispanic white people. We conducted multi-omics analyses to elucidate molecular alterations in HCC among Hispanic patients. METHODS: Paired tumor and adjacent non-tumor samples were collected from 31 Hispanic HCCs in South Texas (STX-Hispanic) for genomic, transcriptomic, proteomic, and metabolomic profiling. Serum lipids were profiled in 40 Hispanic and non-Hispanic patients with or without clinically diagnosed HCC. RESULTS: Exome sequencing revealed high mutation frequencies of AXIN2 and CTNNB1 in STX Hispanic HCCs, suggesting a predominant activation of the Wnt/ß-catenin pathway. TERT promoter mutations were also significantly more frequent in the Hispanic cohort (Fisher's exact test, p < .05). Cell cycles and liver function were positively and negatively enriched, respectively, with gene set enrichment analysis. Gene sets representing specific liver metabolic pathways were associated with dysregulation of corresponding metabolites. Negative enrichment of liver adipogenesis and lipid metabolism corroborated with a significant reduction in most lipids in serum samples of HCC patients (paired t-test, p < .0001). Two HCC subtypes from our Hispanic cohort were identified and validated with the TCGA liver cancer cohort. Patients with better overall survival showed higher activity of immune and angiogenesis signatures, and lower activity of liver function-related gene signatures. They also had higher levels of immune checkpoint and immune exhaustion markers. CONCLUSIONS: Our study revealed specific molecular features of Hispanic HCC and potential biomarkers for therapeutic management. It provides a unique resource for studying Hispanic HCC.

Antitumor Activity of a Novel LAIR1 Antagonist in Combination with Anti-PD1 to Treat Collagen-Rich Solid Tumors.

We recently reported that resistance to PD-1 blockade in a refractory lung cancer-derived model involved increased collagen deposition and the collagen-binding inhibitory receptor leukocyte-associated immunoglobulin-like receptor 1 (LAIR1). Thus, we hypothesized that LAIR1 and collagen cooperated to suppress therapeutic response. In this study, we report that LAIR1 is associated with tumor stroma and is highly expressed by intratumoral myeloid cells in both human tumors and mouse models of cancer. Stroma-associated myeloid cells exhibit a suppressive phenotype and correlate with LAIR1 expression in human cancer. NGM438, a novel humanized LAIR1 antagonist mAb, elicits myeloid inflammation and allogeneic T-cell responses by binding to LAIR1 and blocking collagen engagement. Furthermore, a mouse-reactive NGM438 surrogate antibody sensitized refractory KP mouse lung tumors to anti-PD-1 therapy and resulted in increased intratumoral CD8+ T-cell content and inflammatory gene expression. These data place LAIR1 at the intersection of stroma and suppressive myeloid cells and support the notion that blockade of the LAIR1/collagen axis can potentially address resistance to checkpoint inhibitor therapy in the clinic.

ILT2 and ILT4 Drive Myeloid Suppression via Both Overlapping and Distinct Mechanisms.

Solid tumors are dense three-dimensional (3D) multicellular structures that enable efficient receptor-ligand trans interactions via close cell-cell contact. Immunoglobulin-like transcript (ILT)2 and ILT4 are related immune-suppressive receptors that play a role in the inhibition of myeloid cells within the tumor microenvironment. The relative contribution of ILT2 and ILT4 to immune inhibition in the context of solid tumor tissue has not been fully explored. We present evidence that both ILT2 and ILT4 contribute to myeloid inhibition. We found that although ILT2 inhibits myeloid cell activation in the context of trans-engagement by MHC-I, ILT4 efficiently inhibits myeloid cells in the presence of either cis- or trans-engagement. In a 3D spheroid tumor model, dual ILT2/ILT4 blockade was required for the optimal activation of myeloid cells, including the secretion of CXCL9 and CCL5, upregulation of CD86 on dendritic cells, and downregulation of CD163 on macrophages. Humanized mouse tumor models showed increased immune activation and cytolytic T-cell activity with combined ILT2 and ILT4 blockade, including evidence of the generation of immune niches, which have been shown to correlate with clinical response to immune-checkpoint blockade. In a human tumor explant histoculture system, dual ILT2/ILT4 blockade increased CXCL9 secretion, downregulated CD163 expression, and increased the expression of M1 macrophage, IFNγ, and cytolytic T-cell gene signatures. Thus, we have revealed distinct contributions of ILT2 and ILT4 to myeloid cell biology and provide proof-of-concept data supporting the combined blockade of ILT2 and ILT4 to therapeutically induce optimal myeloid cell reprogramming in the tumor microenvironment.

The Fibronectin-ILT3 Interaction Functions as a Stromal Checkpoint that Suppresses Myeloid Cells.

Suppressive myeloid cells inhibit antitumor immunity by preventing T-cell responses. Immunoglobulin-like transcript 3 (ILT3; also known as LILRB4) is highly expressed on tumor-associated myeloid cells and promotes their suppressive phenotype. However, the ligand that engages ILT3 within the tumor microenvironment and renders tumor-associated myeloid cells suppressive is unknown. Using a screening approach, we identified fibronectin as a functional ligand for ILT3. The interaction of fibronectin with ILT3 polarized myeloid cells toward a suppressive state, and these effects were reversed with an ILT3-specific antibody that blocked the interaction of ILT3 with fibronectin. Furthermore, ex vivo treatment of human tumor explants with anti-ILT3 reprogrammed tumor-associated myeloid cells toward a stimulatory phenotype. Thus, the ILT3-fibronectin interaction represents a "stromal checkpoint" through which the extracellular matrix actively suppresses myeloid cells. By blocking this interaction, tumor-associated myeloid cells may acquire a stimulatory phenotype, potentially resulting in increased antitumor T-cell responses.

Antibody-mediated inhibition of GDF15-GFRAL activity reverses cancer cachexia in mice.

Cancer cachexia is a highly prevalent condition associated with poor quality of life and reduced survival1. Tumor-induced perturbations in the endocrine, immune and nervous systems drive anorexia and catabolic changes in adipose tissue and skeletal muscle, hallmarks of cancer cachexia2-4. However, the molecular mechanisms driving cachexia remain poorly defined, and there are currently no approved drugs for the condition. Elevation in circulating growth differentiation factor 15 (GDF15) correlates with cachexia and reduced survival in patients with cancer5-8, and a GDNF family receptor alpha like (GFRAL)-Ret proto-oncogene (RET) signaling complex in brainstem neurons that mediates GDF15-induced weight loss in mice has recently been described9-12. Here we report a therapeutic antagonistic monoclonal antibody, 3P10, that targets GFRAL and inhibits RET signaling by preventing the GDF15-driven interaction of RET with GFRAL on the cell surface. Treatment with 3P10 reverses excessive lipid oxidation in tumor-bearing mice and prevents cancer cachexia, even under calorie-restricted conditions. Mechanistically, activation of the GFRAL-RET pathway induces expression of genes involved in lipid metabolism in adipose tissues, and both peripheral chemical sympathectomy and loss of adipose triglyceride lipase protect mice from GDF15-induced weight loss. These data uncover a peripheral sympathetic axis by which GDF15 elicits a lipolytic response in adipose tissue independently of anorexia, leading to reduced adipose and muscle mass and function in tumor-bearing mice.

A probe-density-based analysis method for array CGH data: simulation, normalization and centralization.

MOTIVATION: Genomic instability is one of the fundamental factors in tumorigenesis and tumor progression. Many studies have shown that copy-number abnormalities at the DNA level are important in the pathogenesis of cancer. Array comparative genomic hybridization (aCGH), developed based on expression microarray technology, can reveal the chromosomal aberrations in segmental copies at a high resolution. However, due to the nature of aCGH, many standard expression data processing tools, such as data normalization, often fail to yield satisfactory results. RESULTS: We demonstrated a novel aCGH normalization algorithm, which provides an accurate aCGH data normalization by utilizing the dependency of neighboring probe measurements in aCGH experiments. To facilitate the study, we have developed a hidden Markov model (HMM) to simulate a series of aCGH experiments with random DNA copy number alterations that are used to validate the performance of our normalization. In addition, we applied the proposed normalization algorithm to an aCGH study of lung cancer cell lines. By using the proposed algorithm, data quality and the reliability of experimental results are significantly improved, and the distinct patterns of DNA copy number alternations are observed among those lung cancer cell lines. SUPPLEMENTARY INFORMATION: Source codes and.gures may be found at http://ntumaps.cgm.ntu.edu.tw/aCGH_supplementary.

Predicting drug response of tumors from integrated genomic profiles by deep neural networks.

BACKGROUND: The study of high-throughput genomic profiles from a pharmacogenomics viewpoint has provided unprecedented insights into the oncogenic features modulating drug response. A recent study screened for the response of a thousand human cancer cell lines to a wide collection of anti-cancer drugs and illuminated the link between cellular genotypes and vulnerability. However, due to essential differences between cell lines and tumors, to date the translation into predicting drug response in tumors remains challenging. Recently, advances in deep learning have revolutionized bioinformatics and introduced new techniques to the integration of genomic data. Its application on pharmacogenomics may fill the gap between genomics and drug response and improve the prediction of drug response in tumors. RESULTS: We proposed a deep learning model to predict drug response (DeepDR) based on mutation and expression profiles of a cancer cell or a tumor. The model contains three deep neural networks (DNNs), i) a mutation encoder pre-trained using a large pan-cancer dataset (The Cancer Genome Atlas; TCGA) to abstract core representations of high-dimension mutation data, ii) a pre-trained expression encoder, and iii) a drug response predictor network integrating the first two subnetworks. 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 (linear regression and support vector machine) and four analog DNN models of DeepDR, including DNNs built without TCGA pre-training, partly replaced by principal components, and built on individual types of input data. We then applied the model to predict drug response of 9059 tumors of 33 cancer types. Using per-cancer and pan-cancer settings, the model predicted both known, including EGFR inhibitors in non-small cell lung cancer and tamoxifen in ER+ breast cancer, and novel drug targets, such as vinorelbine for TTN-mutated tumors. 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. CONCLUSIONS: Here we present, as far as we know, the first DNN model to translate pharmacogenomics features identified from in vitro drug screening to predict the response of tumors. The results covered both well-studied and novel mechanisms of drug resistance and drug targets. Our model and findings improve the prediction of drug response and the identification of novel therapeutic options.

Correction to: Predicting drug response of tumors from integrated genomic profiles by deep neural networks.

Following publication of the original article [1], the authors provided an updated funding statement to the article. The updated statement is as follows.

GSAE: an autoencoder with embedded gene-set nodes for genomics functional characterization.

BACKGROUND: Bioinformatics tools have been developed to interpret gene expression data at the gene set level, and these gene set based analyses improve the biologists' capability to discover functional relevance of their experiment design. While elucidating gene set individually, inter-gene sets association is rarely taken into consideration. Deep learning, an emerging machine learning technique in computational biology, can be used to generate an unbiased combination of gene set, and to determine the biological relevance and analysis consistency of these combining gene sets by leveraging large genomic data sets. RESULTS: In this study, we proposed a gene superset autoencoder (GSAE), a multi-layer autoencoder model with the incorporation of a priori defined gene sets that retain the crucial biological features in the latent layer. We introduced the concept of the gene superset, an unbiased combination of gene sets with weights trained by the autoencoder, where each node in the latent layer is a superset. Trained with genomic data from TCGA and evaluated with their accompanying clinical parameters, we showed gene supersets' ability of discriminating tumor subtypes and their prognostic capability. We further demonstrated the biological relevance of the top component gene sets in the significant supersets. CONCLUSIONS: Using autoencoder model and gene superset at its latent layer, we demonstrated that gene supersets retain sufficient biological information with respect to tumor subtypes and clinical prognostic significance. Superset also provides high reproducibility on survival analysis and accurate prediction for cancer subtypes.

Utilizing Cancer - Functional Gene Set - Compound Networks to Identify Putative Drugs for Breast Cancer.

AIM AND OBJECTIVE: The number of anticancer drugs available currently is limited, and some of them have low treatment response rates. Moreover, developing a new drug for cancer therapy is labor intensive and sometimes cost prohibitive. Therefore, "repositioning" of known cancer treatment compounds can speed up the development time and potentially increase the response rate of cancer therapy. This study proposes a systems biology method for identifying new compound candidates for cancer treatment in two separate procedures. MATERIALS AND METHODS: First, a "gene set-compound" network was constructed by conducting gene set enrichment analysis on the expression profile of responses to a compound. Second, survival analyses were applied to gene expression profiles derived from four breast cancer patient cohorts to identify gene sets that are associated with cancer survival. A "cancer-functional gene set- compound" network was constructed, and candidate anticancer compounds were identified. Through the use of breast cancer as an example, 162 breast cancer survival-associated gene sets and 172 putative compounds were obtained. RESULTS: We demonstrated how to utilize the clinical relevance of previous studies through gene sets and then connect it to candidate compounds by using gene expression data from the Connectivity Map. Specifically, we chose a gene set derived from a stem cell study to demonstrate its association with breast cancer prognosis and discussed six new compounds that can increase the expression of the gene set after the treatment. CONCLUSION: Our method can effectively identify compounds with a potential to be "repositioned" for cancer treatment according to their active mechanisms and their association with patients' survival time.

IGFBP7 Deletion Promotes Hepatocellular Carcinoma.

Activation of IGF signaling is a major oncogenic event in diverse cancers, including hepatocellular carcinoma (HCC). In this setting, the insulin-like growth factor binding protein IGFBP7 inhibits IGF signaling by binding the IGF1 receptor (IGF1R), functioning as a candidate tumor suppressor. IGFBP7 abrogates tumors by inhibiting angiogenesis and inducing cancer-specific senescence and apoptosis. Here, we report that Igfbp7-deficient mice exhibit constitutively active IGF signaling, presenting with proinflammatory and immunosuppressive microenvironments and spontaneous liver and lung tumors occurring with increased incidence in carcinogen-treated subjects. Igfbp7 deletion increased proliferation and decreased senescence of hepatocytes and mouse embryonic fibroblasts, effects that were blocked by treatment with IGF1 receptor inhibitor. Significant inhibition of genes regulating immune surveillance was observed in Igfbp7-/- murine livers, which was associated with a marked inhibition in antigen cross-presentation by Igfbp7-/- dendritic cells. Conversely, IGFBP7 overexpression inhibited growth of HCC cells in syngeneic immunocompetent mice. Depletion of CD4+ or CD8+ T lymphocytes abolished this growth inhibition, identifying it as an immune-mediated response. Our findings define an immune component of the pleiotropic mechanisms through which IGFBP7 suppresses HCC. Furthermore, they offer a genetically based preclinical proof of concept for IGFBP7 as a therapeutic target for immune management of HCC. Cancer Res; 77(15); 4014-25. ©2017 AACR.