Identification of early liver toxicity gene biomarkers using comparative supervised machine learning

Screening agrochemicals and pharmaceuticals for potential liver toxicity is required for regulatory approval and is an expensive and time-consuming process. The identification and utilization of early exposure gene signatures and robust predictive models in regulatory toxicity testing has the potential to reduce time and costs substantially. In this study, comparative supervised machine learning approaches were applied to the rat liver TG-GATEs dataset to develop feature selection and predictive testing. We identified ten gene biomarkers using three different feature selection methods that predicted liver necrosis with high specificity and selectivity in an independent validation dataset from the Microarray Quality Control (MAQC)-II study. Nine of the ten genes that were selected with the supervised methods are involved in metabolism and detoxification (Car3, Crat, Cyp39a1, Dcd, Lbp, Scly, Slc23a1, and Tkfc) and transcriptional regulation (Ablim3). Several of these genes are also implicated in liver carcinogenesis, including Crat, Car3 and Slc23a1. Our biomarker gene signature provides high statistical accuracy and a manageable number of genes to study as indicators to potentially accelerate toxicity testing based on their ability to induce liver necrosis and, eventually, liver cancer.

PolyA-miner: accurate assessment of differential alternative poly-adenylation from 3′Seq data using vector projections and non-negative matrix factorization

Almost 70% of human genes undergo alternative polyadenylation (APA) and generate mRNA transcripts with varying lengths, typically of the 3′ untranslated regions (UTR). APA plays an important role in development and cellular differentiation, and its dysregulation can cause neuropsychiatric diseases and increase cancer severity. Increasing awareness of APA’s role in human health and disease has propelled the development of several 3′ sequencing (3′Seq) techniques that allow for precise identification of APA sites. However, despite the recent data explosion, there are no robust computational tools that are precisely designed to analyze 3′Seq data. Analytical approaches that have been used to analyze these data predominantly use proximal to distal usage. With about 50% of human genes having more than two APA isoforms, current methods fail to capture the entirety of APA changes and do not account for non-proximal to non-distal changes. Addressing these key challenges, this study demonstrates PolyA-miner, an algorithm to accurately detect and assess differential alternative polyadenylation specifically from 3′Seq data. Genes are abstracted as APA matrices, and differential APA usage is inferred using iterative consensus non-negative matrix factorization (NMF) based clustering. PolyA-miner accounts for all non-proximal to non-distal APA switches using vector projections and reflects precise gene-level 3′UTR changes. It can also effectively identify novel APA sites that are otherwise undetected when using reference-based approaches. Evaluation on multiple datasets—first-generation MicroArray Quality Control (MAQC) brain and Universal Human Reference (UHR) PolyA-seq data, recent glioblastoma cell line NUDT21 knockdown Poly(A)-ClickSeq (PAC-seq) data, and our own mouse hippocampal and human stem cell-derived neuron PAC-seq data—strongly supports the value and protocol-independent applicability of PolyA-miner. Strikingly, in the glioblastoma cell line data, PolyA-miner identified more than twice the number of genes with APA changes than initially reported. With the emerging importance of APA in human development and disease, PolyA-miner can significantly improve data analysis and help decode the underlying APA dynamics.

Mixture models reveal multiple positional bias types in RNA-Seq data and lead to accurate transcript concentration estimates

Quantification of RNA transcripts with RNA-Seq is inaccurate due to positional fragment bias, which is not represented appropriately by current statistical models of RNA-Seq data. This article introduces the Mix2(rd. "mixquare") model, which uses a mixture of probability distributions to model the transcript specific positional fragment bias. The parameters of the Mix2model can be efficiently trained with the Expectation Maximization (EM) algorithm resulting in simultaneous estimates of the transcript abundances and transcript specific positional biases. Experiments are conducted on synthetic data and the Universal Human Reference (UHR) and Brain (HBR) sample from the Microarray quality control (MAQC) data set. Comparing the correlation between qPCR and FPKM values to state-of-the-art methods Cufflinks and PennSeq we obtain an increase in R2value from 0.44 to 0.6 and from 0.34 to 0.54. In the detection of differential expression between UHR and HBR the true positive rate increases from 0.44 to 0.71 at a false positive rate of 0.1. Finally, the Mix2model is used to investigate biases present in the MAQC data. This reveals 5 dominant biases which deviate from the common assumption of a uniform fragment distribution. The Mix2software is available at http://www.lexogen.com/fileadmin/uploads/bioinfo/mix2model.tgz.

Comparing Imputation Procedures for Affymetrix Gene Expression Datasets Using MAQC Datasets

Introduction. The microarray datasets from the MicroArray Quality Control (MAQC) project have enabled the assessment of the precision, comparability of microarrays, and other various microarray analysis methods. However, to date no studies that we are aware of have reported the performance of missing value imputation schemes on the MAQC datasets. In this study, we use the MAQC Affymetrix datasets to evaluate several imputation procedures in Affymetrix microarrays. Results. We evaluated several cutting edge imputation procedures and compared them using different error measures. We randomly deleted 5% and 10% of the data and imputed the missing values using imputation tests. We performed 1000 simulations and averaged the results. The results for both 5% and 10% deletion are similar. Among the imputation methods, we observe the local least squares method with is most accurate under the error measures considered. The k-nearest neighbor method with has the highest error rate among imputation methods and error measures. Conclusions. We conclude for imputing missing values in Affymetrix microarray datasets, using the MAS 5.0 preprocessing scheme, the local least squares method with has the best overall performance and k-nearest neighbor method with has the worst overall performance. These results hold true for both 5% and 10% missing values.