Biology subjects

Badsha, M. B.

Publications and source records attributed to Badsha, M. B..

4 recordsLinked to original sources

bioRxiv2022

Types of cis- and trans-gene regulation of expression quantitative trait loci across human tissues

Expression quantitative trait loci (eQTLs) have been identified for most genes in the human genome across tissues and cell types. While most of the eQTLs are near the associated genes, some can be far away or on different chromosomes, with the regulatory mechanisms largely unknown. Here, we study cis- and trans-regulation by eQTLs on protein-coding genes and long noncoding RNAs (lncRNAs) across nearly 50 tissues and cell types. Specifically, we constructed trios consisting of an eQTL, its cis-gene and trans-gene and inferred the regulatory relationships with causal network inference. We identify multiple types of regulatory networks for trios: across all the tissues, more than half of the trios are inferred to be conditionally independent, where the two genes are conditionally independent given the genotype of the eQTL (gene 1 [<-] eQTL [->] gene 2). Around 1.5% of the trios are inferred to be mediation (eQTL [->] mediator [->] target), around 1.3% fully connected among the three nodes, and just a handful v-structures (eQTL [->] gene 1 [<-] gene 2). Unexpectedly, across the tissues, on average more than half of the mediation trios have the trans-gene as the mediator. Most of the mediators (cis and trans) are tissue specific, and cis-gene mediators are significantly enriched for protein-coding genes, whereas trans-gene mediators have a similar distribution of protein-coding genes and lncRNAs to the whole genome.

genomics↗

bioRxiv2020

Robust Hierarchical Co-clustering to Explore Toxicogenomic Biomarkers and Their Regulatory Doses of Chemical Compounds

Toxicogenomics combines high throughput molecular technologies with statistical and machine learning approaches to discover a similar group of doses of chemical compounds (DCCs) and genes to explore toxicogenomic biomarkers and their regulatory DCCs. This is also very important in the toxicity study of environmental stressors, synthetic chemicals and drug discovery and development process. Different clustering algorithms are concerned with the discovering of interesting clusters/groups of row or column entities of a dataset. Among those hierarchical clustering (HC) and logistic probabilistic hidden variable model (LPHVM) can identify toxicogenomic biomarkers and their regulatory DCCs forming co-cluster. However, the HC method is very sensitive to outlying observations. On the other hand, though LPHVM is a robust approach, it consumes more time for calculation since it is Expectation-Maximization (EM) based iterative approach. Additionally, the LPHVM creates artificiality problem taking absolute value of the data matrix. Therefore, to overcome these problems in this paper, we proposed a robust hierarchical co-clustering (RHCOC) algorithm to co-cluster genes and DCCs simultaneously with a view to explore toxicogenomic biomarkers and their regulatory DCCs. The performance of the proposed RHCOC algorithm over the conventional HC for clustering genes and DCCs of toxicogenomic data has been investigated based on the simulation study. The results of the simulation study have shown that the RHCOC approaches produce far lower clustering error rate (ER) than the conventional HC approaches in presence of outlying observations in the dataset. Otherwise they perform equally in absence of outlier in the dataset. To explore biomarker co-clusters consisting of toxicogenomic biomarker genes and their regulatory DCCs we used control chart for individual measurement (CCIM). We have also investigated the performance of the proposed approach in the case of the pathway level real life fold change gene expression (FCGE) toxicogenomic data analysis. The biomarker co-clusters consisting of toxicogenomic biomarker genes and their regulatory DCCs and biomarker genes explored by the proposed approaches have been validated by the literature and functional annotation. Our method is implemented in R package "rhcoclust" available on github (https://github.com/mdbahadur/rhcoclust).

bioinformatics↗

bioRxiv2018

Imputation of single-cell gene expression with an autoencoder neural network

BackgroundSingle-cell RNA-sequencing (scRNA-seq) is a rapidly evolving technology that enables measurement of gene expression levels at an unprecedented resolution. Despite the explosive growth in the number of cells that can be assayed by a single experiment, scRNA-seq still has several limitations, including high rates of dropouts, which result in a large number of genes having zero read count in the scRNA-seq data, and complicate downstream analyses.\n\nMethodsTo overcome this problem, we treat zeros as missing values and develop nonparametric deep learning methods for imputation. Specifically, our LATE (Learning with AuToEncoder) method trains an autoencoder with random initial values of the parameters, whereas our TRANSLATE (TRANSfer learning with LATE) method further allows for the use of a reference gene expression data set to provide LATE with an initial set of parameter estimates.\n\nResultsOn both simulated and real data, LATE and TRANSLATE outperform existing scRNA-seq imputation methods, achieving lower mean squared error in most cases, recovering nonlinear gene-gene relationships, and better separating cell types. They are also highly scalable and can efficiently process over 1 million cells in just a few hours on a GPU.\n\nConclusionsWe demonstrate that our nonparametric approach to imputation based on autoencoders is powerful and highly efficient.

bioinformatics↗

bioRxiv2017

Learning causal biological networks with generalized Mendelian randomization

Although large amounts of genomic data are available, it remains a challenge to reliably infer causal (i.e., regulatory) relationships among molecular phenotypes (such as gene expression), especially when many phenotypes are involved. We extend the interpretation of the Principle of Mendelian randomization (PMR) and present MRPC, a novel machine learning algorithm that incorporates the PMR in classical algorithms for learning causal graphs in computer science. MRPC learns a causal biological network efficiently and robustly from integrating genotype and molecular phenotype data, in which directed edges indicate causal directions. We demonstrate through simulation that MRPC outperforms existing general-purpose network inference methods and other PMR-based methods. We apply MRPC to distinguish direct and indirect targets among multiple genes associated with expression quantitative trait loci.

bioinformatics↗