RETRACTED: Identifying halophilic proteins based on random forests with preprocessing of the pseudo-amino acid composition

Investigating genetic characteristics of hepatitis B virus-associated and -non-associated hepatocellular carcinoma

BACKGROUND

Hepatocellular carcinoma (HCC) is a primary liver malignancy that mainly occurs in patients with chronic liver disease and cirrhosis. Risk factors for HCC include hepatitis B virus (HBV) infection. However, the specific role of HBV infection in HCC development is not yet completely understood. In order to reveal the effects of HBV on HCC, we compare the genes of HCC patients infected with HBV with those who are not infected.

METHODS

We encoded the genes of these two types of HCC in databases using enrichment scores of Gene Ontology and Kyoto Encyclopedia of Genes and Genomes pathway terms. A random forest algorithm was employed in order to distinguish these two types in the classifier, and a series of feature selection approaches was used in order to select their optimal features. Novel HBV-associated and -non-associated HCC genes were predicted, respectively, based on their optimal features in the classifier. A shortest-path algorithm was also employed in order to find all of the shortest-paths genes connecting the known related genes.

RESULTS

A total of 54 different features between HBV-associated and -non-associated HCC genes were identified. In total, 1236 and 881 novel related genes were predicted for HBV-associated and -non-associated HCC, respectively. By integrating the predicted genes and shortest path genes in their gene interaction network, we identified 679 common genes involved in the two types of HCC.

CONCLUSION

We identified the significantly different genetic features between two types of HCC. We also predicted related genes for the two types based on their specific features. Finally, we determined the common genes and features that were involved in both of these two types of HCC.

Predicting novel salivary biomarkers for the detection of pancreatic cancer using biological feature-based classification

AIM

The use of saliva as a diagnostic fluid enables non-invasive sampling and thus is a prospective sample for disease tests. This study fully utilized the information from the salivary transcriptome to characterize pancreatic cancer related genes and predict novel salivary biomarkers.

METHODS

We calculated the enrichment scores of gene ontology (GO) and pathways annotated in Kyoto Encyclopedia of Genes and Genomes database (KEGG) for pancreatic cancer-related genes. Annotation of GO and KEGG pathway characterize the molecular features of genes. We employed Random Forest classification and incremental feature selection to identify the optimal features among them and predicted novel pancreatic cancer-related genes.

RESULTS

A total of 2175 gene ontology and 79 KEGG pathway terms were identified as the optimal features to identify pancreatic cancer-related genes. A total of 516 novel genes were predicted using these features. We discovered 29 novel biomarkers based on the expression of these 516 genes in saliva. Using our new biomarkers, we achieved a higher accuracy (92%) for the detection of pancreatic cancer. Another independent expression dataset confirmed that these novel biomarkers performed better than the previously described markers alone.

CONCLUSION

By analyzing the information of the salivary transcriptome, we predict pancreatic cancer-related genes and novel salivary gene markers for detection.

Finding gastric cancer related genes and clinical biomarkers for detection based on gene-gene interaction network

BACKGROUND/OBJECTIVE

Gastric cancer (GC) is the second leading cause of death resulted from cancer globally. The most common cause of GC is the infection of Helicobacter pylori, approximately 11% of cases are caused by genetic factors. The objective of this study was to develop an effective computational method to meaningfully interpret these GC-related genes and to predict potential prognostic genes for clinical detection.

METHODS

We employed the shortest path algorithm and permutation test to probe the genes that have relationship with known GC genes in gene-gene interaction network. We calculated the enrichment scores of gene ontology and pathways of gastric cancer related genes to characterize these genes in terms of molecular features. The optimal features that primly representing the gastric cancer related genes were selected using Random Forest classification and incremental feature selection. Random Forest classification was also used for the prediction of the novel gastric cancer related genes based on the selected features and the identification of novel prognostic genes based on the expression of genes.

RESULTS

Based on the shortest path analysis of 36 known GC genes, 39 genes occurring in shortest path were identified as GC-related genes. In subsequent classification, 4153 gene ontology terms and 157 pathway terms were identified as the optimal features to depict these gastric cancer related genes. Based on them, a total of 886 genes were predicted as related genes. These 886 genes could serve as expression biomarkers for clinical detection and they achieved a 100% accuracy for distinguishing gastric cancer from a case-control dataset, better than any of 886 random selected genes did.

CONCLUSION

By analyzing the features of known GC-related genes, we employed a systematic method to predict gastric cancer related genes and novel prognostic genes for accurate clinical detection.

Interpreting the distinct and shared genetic characteristics between Epstein-Barr virus associated and non-associated gastric carcinoma

Gastric carcinoma is one of the major causes of cancer mortality worldwide. There is a better prognosis for patients with Epstein-Barr virus (EBV)-associated gastric carcinoma (EBVaGC) compared with those with EBV negative gastric carcinoma (EBVnGC). It is partly due to the fact that EBV infection recruits lymphocytes infiltrating the tumor. It has been reported that this infection indeed resulted in the changes in immune response genes and thus preventing the development of tumor. It is worthwhile to do a systematic study of EBVaGC and EBVnGC based on genetic characteristics and pathways. In this study, we investigated the information of gene ontology (GO) and KEGG pathway annotations to characterize EBVaGC and EBVnGC-related genes. By applying minimum redundancy maximum relevance (mRMR) algorithm, we provided an optimal set of features for identifying the EBVaGC and EBVnGC. We also employed the shortest path algorithm to probe the novel EBVaGC- and EBVnGC-related genes based on the interaction network of genes that differently expressed in them respectively. We obtained 1039 and 1003 features to identify these two types of gastric carcinoma respectively. Based on the optimal features of classification, we predicted 1881 and 2475 novel genes as additional candidates to support clinical research respectively for these two types of gastric cancers. We compared the differences and similarities of molecular traits between EBVaGC and EBVnGC, which would facilitate the understanding of gastric cancer and its therapy and was thus clinically relevant.

Predicting A-to-I RNA editing by feature selection and random forest

RNA editing is a post-transcriptional RNA process that provides RNA and protein complexity for regulating gene expression in eukaryotes. It is challenging to predict RNA editing by computational methods. In this study, we developed a novel method to predict RNA editing based on a random forest method. A careful feature selection procedure was performed based on the Maximum Relevance Minimum Redundancy (mRMR) and Incremental Feature Selection (IFS) algorithms. Eighteen optimal features were selected from the 77 features in our dataset and used to construct a final predictor. The accuracy and MCC (Matthews correlation coefficient) values for the training dataset were 0.866 and 0.742, respectively; for the testing dataset, the accuracy and MCC were 0.876 and 0.576, respectively. The performance was higher using 18 features than all 77, suggesting that a small feature set was sufficient to achieve accurate prediction. Analysis of the 18 features was performed and may shed light on the mechanism and dominant factors of RNA editing, providing a basis for future experimental validation.