Search

BackgroundDrug sensitivity prediction and drug responsive biomarker selection on high-throughput genomic data is a critical step in drug discovery. Many computational methods have been developed to serve this purpose including several deep neural network models. However, the modular relations among genomic features have been largely ignored in these methods. To overcome this limitation, the role of the gene co-expression network on drug sensitivity prediction is investigated in this study.MethodsIn this paper, we first introduce a network-based method to identify representative features for drug response prediction by using the gene co-expression network. Then, two graph-based neural network models are proposed and both models integrate gene network information directly into neural network for outcome prediction. Next, we present a large-scale comparative study among the proposed network-based methods, canonical prediction algorithms (i.e., Elastic Net, Random Forest, Partial Least Squares Regression, and Support Vector Regression), and deep neural network models for drug sensitivity prediction. All the source code and processed datasets in this study are available athttps://github.com/compbiolabucf/drug-sensitivity-prediction.ResultsIn the comparison of different feature selection methods and prediction methods on a non-small cell lung cancer (NSCLC) cell line RNA-seq gene expression dataset with 50 different drug treatments, we found that (1) the network-based feature selection method improves the prediction performance compared to Pearson correlation coefficients; (2) Random Forest outperforms all the other canonical prediction algorithms and deep neural network models; (3) the proposed graph-based neural network models show better prediction performance compared to deep neural network model; (4) the prediction performance is drug dependent and it may relate to the drug’s mechanism of action.ConclusionsNetwork-based feature selection method and prediction models improve the performance of the drug response prediction. The relations between the genomic features are more robust and stable compared to the correlation between each individual genomic feature and the drug response in high dimension and low sample size genomic datasets.

The pathogenesis of ulcerative colitis (UC), a major type of inflammatory bowel disease, remains unknown. No model exists that adequately recapitulates the complexity of clinical UC. Here, we take advantage of induced pluripotent stem cells (iPSCs) to develop an induced human UC-derived organoid (iHUCO) model and compared it with the induced human normal organoid model (iHNO). Notably, iHUCOs recapitulated histological and functional features of primary colitic tissues, including the absence of acidic mucus secretion and aberrant adherens junctions in the epithelial barrier both in vitro and in vivo. We demonstrate that the CXCL8/CXCR1 axis was overexpressed in iHUCO but not in iHNO. As proof-of-principle, we show that inhibition of CXCL8 receptor by the small-molecule non-competitive inhibitor repertaxin attenuated the progression of UC phenotypes in vitro and in vivo. This patient-derived organoid model, containing both epithelial and stromal compartments, will generate new insights into the underlying pathogenesis of UC while offering opportunities to tailor interventions to the individual patient., Although ulcerative colitis (UC) is a major type of inflammatory bowel disease, attempts to model it fully have fallen short. Here the authors use patient-derived iPS cells to develop a UC organoid model that recapitulates disease histological and functional features, and confirm the role of CXCL8/CXCR1 in pathogenesis.

Hispanic/Latino patients have a higher incidence of gastric cancer and worse cancer-related outcomes compared to patients of other backgrounds. Whether there is a molecular basis for these disparities is unknown, as very few Hispanic/Latino patients have been included in previous studies. To determine the genomic landscape of gastric cancer in Hispanic/Latino patients, we performed whole-exome sequencing (WES) and RNA sequencing on tumor samples from 57 patients; germline analysis was conducted on 83 patients. The results were compared to data from Asian and White patients published by The Cancer Genome Atlas. Hispanic/Latino patients had a significantly larger proportion of genomically-stable subtype tumors compared to Asian and White patients (65% vs 21% vs 20%, P < 0.001). Transcriptomic analysis identified molecular signatures that were prognostic. Of the 43 Hispanic/Latino patients with diffuse-type cancer, 7 (16%) had germline mutations in CDH1. Mutation carriers were significantly younger than non-carriers (41 vs 50 years, P < 0.05). In silico algorithms predicted 5 variants to be deleterious. For two variants that were predicted to be benign, in vitro modeling demonstrated that these mutations conferred increased migratory capability, suggesting pathogenicity. Hispanic/Latino gastric cancer patients possess unique genomic landscapes, including a high rate of CDH1 germline mutations that may partially explain their aggressive clinical phenotypes. Individualized screening, genetic counseling, and treatment protocols based on patient ethnicity and race may be necessary.

The introduction of insertion-deletions (INDELs) by non-homologous end-joining (NHEJ) pathway underlies the mechanistic basis of CRISPR-Cas9-directed genome editing. Selective gene ablation using CRISPR-Cas9 is achieved by installation of a premature termination codon (PTC) from a frameshift-inducing INDEL that elicits nonsense-mediated decay (NMD) of the mutant mRNA. Here, by examining the mRNA and protein products of CRISPR targeted genes in a cell line panel with presumed gene knockouts, we detect the production of foreign mRNAs or proteins in ~50% of the cell lines. We demonstrate that these aberrant protein products stem from the introduction of INDELs that promote internal ribosomal entry, convert pseudo-mRNAs (alternatively spliced mRNAs with a PTC) into protein encoding molecules, or induce exon skipping by disruption of exon splicing enhancers (ESEs). Our results reveal challenges to manipulating gene expression outcomes using INDEL-based mutagenesis and strategies useful in mitigating their impact on intended genome-editing outcomes., CRISPR-Cas9 genome editing is presumed to knock out gene function by generating a frameshift during NHEJ repair. Here, the authors investigate mRNA and protein expression in edited lines and find genome editing can generate internal ribosome entry sites or alternatively spliced variants.

Identifying alternative indications for known drugs is important for the pharmaceutical industry. Many computational methods have been proposed for predicting unknown associations between drugs and target proteins associated with diseases. To produce better prediction, researchers should not only develop accurate algorithms but identify good features that reflect intracellular systems. In this paper, we proposed a novel method for exploiting protein localization. We generated localization vectors (LVs) from protein localization and propagated LVs through a protein interaction network to increase the coverage of the localization information. The LVs showed distinct patterns among targets of known drugs as well as independent characteristics compared to existing features. Based on the experimental results, we determined that including LVs improves cross-validation accuracy and, produces better novel predictions with real and independent clinical trial data. Moreover, the propagation of LVs showed a positive result that it can help in increasing the coverage of the prediction results.

There has been much active research in bioinformatics to support our understanding of oncogenesis and tumor progression. Most research relies on mRNA gene expression data to identify marker genes or cancer specific gene networks. However, considering that proteins are functional molecules that carry out the biological tasks of genes, they can be direct markers of biological functions. Protein abundance data on a genome scale have not been investigated in depth due to the limited availability of high throughput protein assays. This hindrance is chiefly caused by a lack of robust techniques such as RT-PCR (real-time polymerase chain reaction). In this study, we quantified phospho-proteomes of breast cancer cell lines treated with TGF-beta (transforming growth factor beta). To discover biomarkers and observe changes in the signaling pathways related to breast cancer, we applied a protein network-based approach to generate a classifier of subnet markers. The accuracy of that classifier outperformed other network-based classification algorithms, and current feature selection and classification algorithms. Moreover, many cancer-related proteins were identified in those sub-networks. Each sub-network provides functional insights and can serve as a potential marker for TGF-beta treatments. After interpreting the roles of proteins in sub-networks with various signaling pathways, we found strong candidate proteins and various related interactions that are expected to affect breast cancer outcomes. These results demonstrate the high quality of the quantified phospho-proteomes data and show that our network construction and classification method is appropriate for an analysis of this type of data.

Background Detecting protein complexes is one of essential and fundamental tasks in understanding various biological functions or processes. Therefore accurate identification of protein complexes is indispensable. Methods For more accurate detection of protein complexes, we propose an algorithm which detects dense protein sub-networks of which proteins share closely located bottleneck proteins. The proposed algorithm is capable of finding protein complexes which allow overlapping with each other. Results We applied our algorithm to several PPI (Protein-Protein Interaction) networks of Saccharomyces cerevisiae and Homo sapiens, and validated our results using public databases of protein complexes. The prediction accuracy was even more improved over our previous work which used also bottleneck information of the PPI network, but showed limitation when predicting small-sized protein complex detection. Conclusions Our algorithm resulted in overlapping protein complexes with significantly improved F1 score over existing algorithms. This result comes from high recall due to effective network search, as well as high precision due to proper use of bottleneck information during the network search.