Comprehensive evaluation of deconvolution methods for human brain gene expression

imply: improving cell-type deconvolution accuracy using personalized reference profiles

Real-world clinical samples are often admixtures of signal mosaics from multiple pure cell types. Using computational tools, bulk transcriptomics can be deconvoluted to solve for the abundance of constituent cell types. However, existing deconvolution methods are conditioned on the assumption that the whole study population is served by a single reference panel, which ignores person-to-person heterogeneity. Here we present imply, a novel algorithm to deconvolute cell type proportions using personalized reference panels. imply can borrow information across repeatedly measured samples for each subject, and obtain precise cell type proportion estimations. Simulation studies demonstrate reduced bias in cell type abundance estimation compared with existing methods. Real data analyses on large longitudinal consortia show more realistic deconvolution results that align with biological facts. Our results suggest that disparities in cell type proportions are associated with several disease phenotypes in type 1 diabetes and Parkinson's disease. Our proposed tool imply is available through the R/Bioconductor package ISLET at https://bioconductor.org/packages/ISLET/.

Identifying cellular markers of focal cortical dysplasia type II with cell-type deconvolution and single-cell signatures

Focal cortical dysplasia (FCD) is a brain malformation that causes medically refractory epilepsy. FCD is classified into three categories based on structural and cellular abnormalities, with FCD type II being the most common and characterized by disrupted organization of the cortex and abnormal neuronal development. In this study, we employed cell-type deconvolution and single-cell signatures to analyze bulk RNA-seq from multiple transcriptomic studies, aiming to characterize the cellular composition of brain lesions in patients with FCD IIa and IIb subtypes. Our deconvolution analyses revealed specific cellular changes in FCD IIb, including neuronal loss and an increase in reactive astrocytes (astrogliosis) when compared to FCD IIa. Astrogliosis in FCD IIb was further supported by a gene signature analysis and histologically confirmed by glial fibrillary acidic protein (GFAP) immunostaining. Overall, our findings demonstrate that FCD II subtypes exhibit differential neuronal and glial compositions, with astrogliosis emerging as a hallmark of FCD IIb. These observations, validated in independent patient cohorts and confirmed using immunohistochemistry, offer novel insights into the involvement of glial cells in FCD type II pathophysiology and may contribute to the development of targeted therapies for this condition.

Fine-grained cell-type specific association studies with human bulk brain data using a large single-nucleus RNA sequencing based reference panel

Brain disorders are leading causes of disability worldwide. Gene expression studies provide promising opportunities to better understand their etiology but it is critical that expression is studied on a cell-type level. Cell-type specific association studies can be performed with bulk expression data using statistical methods that capitalize on cell-type proportions estimated with the help of a reference panel. To create a fine-grained reference panel for the human prefrontal cortex, we performed an integrated analysis of the seven largest single nucleus RNA-seq studies. Our panel included 17 cell-types that were robustly detected across all studies, subregions of the prefrontal cortex, and sex and age groups. To estimate the cell-type proportions, we used an empirical Bayes estimator that substantially outperformed three estimators recommended previously after a comprehensive evaluation of methods to estimate cell-type proportions from brain transcriptome data. This is important as being able to precisely estimate the cell-type proportions may avoid unreliable results in downstream analyses particularly for the multiple cell-types that had low abundances. Transcriptome-wide association studies performed with permuted bulk expression data showed that it is possible to perform transcriptome-wide association studies for even the rarest cell-types without an increased risk of false positives.

BEDwARS: a robust Bayesian approach to bulk gene expression deconvolution with noisy reference signatures

Differential gene expression in bulk transcriptomics data can reflect change of transcript abundance within a cell type and/or change in the proportions of cell types. Expression deconvolution methods can help differentiate these scenarios. BEDwARS is a Bayesian deconvolution method designed to address differences between reference signatures of cell types and corresponding true signatures underlying bulk transcriptomic profiles. BEDwARS is more robust to noisy reference signatures and outperforms leading in-class methods for estimating cell type proportions and signatures. Application of BEDwARS to dihydropyridine dehydrogenase deficiency identified the possible involvement of ciliopathy and impaired translational control in the etiology of the disorder.

Effective methods for bulk RNA-seq deconvolution using scnRNA-seq transcriptomes

BACKGROUND

RNA profiling technologies at single-cell resolutions, including single-cell and single-nuclei RNA sequencing (scRNA-seq and snRNA-seq, scnRNA-seq for short), can help characterize the composition of tissues and reveal cells that influence key functions in both healthy and disease tissues. However, the use of these technologies is operationally challenging because of high costs and stringent sample-collection requirements. Computational deconvolution methods that infer the composition of bulk-profiled samples using scnRNA-seq-characterized cell types can broaden scnRNA-seq applications, but their effectiveness remains controversial.

RESULTS

We produced the first systematic evaluation of deconvolution methods on datasets with either known or scnRNA-seq-estimated compositions. Our analyses revealed biases that are common to scnRNA-seq 10X Genomics assays and illustrated the importance of accurate and properly controlled data preprocessing and method selection and optimization. Moreover, our results suggested that concurrent RNA-seq and scnRNA-seq profiles can help improve the accuracy of both scnRNA-seq preprocessing and the deconvolution methods that employ them. Indeed, our proposed method, Single-cell RNA Quantity Informed Deconvolution (SQUID), which combines RNA-seq transformation and dampened weighted least-squares deconvolution approaches, consistently outperformed other methods in predicting the composition of cell mixtures and tissue samples.

CONCLUSIONS

We showed that analysis of concurrent RNA-seq and scnRNA-seq profiles with SQUID can produce accurate cell-type abundance estimates and that this accuracy improvement was necessary for identifying outcomes-predictive cancer cell subclones in pediatric acute myeloid leukemia and neuroblastoma datasets. These results suggest that deconvolution accuracy improvements are vital to enabling its applications in the life sciences.

Comparative Analysis of Cell Mixtures Deconvolution and Gene Signatures Generated for Blood, Immune and Cancer Cells

In the last two decades, many detailed full transcriptomic studies on complex biological samples have been published and included in large gene expression repositories. These studies primarily provide a bulk expression signal for each sample, including multiple cell-types mixed within the global signal. The cellular heterogeneity in these mixtures does not allow the activity of specific genes in specific cell types to be identified. Therefore, inferring relative cellular composition is a very powerful tool to achieve a more accurate molecular profiling of complex biological samples. In recent decades, computational techniques have been developed to solve this problem by applying deconvolution methods, designed to decompose cell mixtures into their cellular components and calculate the relative proportions of these elements. Some of them only calculate the cell proportions (supervised methods), while other deconvolution algorithms can also identify the gene signatures specific for each cell type (unsupervised methods). In these work, five deconvolution methods (CIBERSORT, FARDEEP, DECONICA, LINSEED and ABIS) were implemented and used to analyze blood and immune cells, and also cancer cells, in complex mixture samples (using three bulk expression datasets). Our study provides three analytical tools (corrplots, cell-signature plots and bar-mixture plots) that allow a thorough comparative analysis of the cell mixture data. The work indicates that CIBERSORT is a robust method optimized for the identification of immune cell-types, but not as efficient in the identification of cancer cells. We also found that LINSEED is a very powerful unsupervised method that provides precise and specific gene signatures for each of the main immune cell types tested: neutrophils and monocytes (of the myeloid lineage), B-cells, NK cells and T-cells (of the lymphoid lineage), and also for cancer cells.

Not every estimate counts - evaluation of cell composition estimation approaches in brain bulk tissue data

BACKGROUND

Variation in cell composition can dramatically impact analyses in bulk tissue samples. A commonly employed approach to mitigate this issue is to adjust statistical models using estimates of cell abundance derived directly from omics data. While an arsenal of estimation methods exists, the applicability of these methods to brain tissue data and whether or not cell estimates can sufficiently account for confounding cellular composition has not been adequately assessed.

METHODS

We assessed the correspondence between different estimation methods based on transcriptomic (RNA sequencing, RNA-seq) and epigenomic (DNA methylation and histone acetylation) data from brain tissue samples of 49 individuals. We further evaluated the impact of different estimation approaches on the analysis of H3K27 acetylation chromatin immunoprecipitation sequencing (ChIP-seq) data from entorhinal cortex of individuals with Alzheimer's disease and controls.

RESULTS

We show that even closely adjacent tissue samples from the same Brodmann area vary greatly in their cell composition. Comparison across different estimation methods indicates that while different estimation methods applied to the same data produce highly similar outcomes, there is a surprisingly low concordance between estimates based on different omics data modalities. Alarmingly, we show that cell type estimates may not always sufficiently account for confounding variation in cell composition.

CONCLUSIONS

Our work indicates that cell composition estimation or direct quantification in one tissue sample should not be used as a proxy to the cellular composition of another tissue sample from the same brain region of an individual-even if the samples are directly adjacent. The highly similar outcomes observed among vastly different estimation methods, highlight the need for brain benchmark datasets and better validation approaches. Finally, unless validated through complementary experiments, the interpretation of analyses outcomes based on data confounded by cell composition should be done with great caution, and ideally avoided all together.

Challenges and opportunities to computationally deconvolve heterogeneous tissue with varying cell sizes using single cell RNA-sequencing datasets

Deconvolution of cell mixtures in "bulk" transcriptomic samples from homogenate human tissue is important for understanding the pathologies of diseases. However, several experimental and computational challenges remain in developing and implementing transcriptomics-based deconvolution approaches, especially those using a single cell/nuclei RNA-seq reference atlas, which are becoming rapidly available across many tissues. Notably, deconvolution algorithms are frequently developed using samples from tissues with similar cell sizes. However, brain tissue or immune cell populations have cell types with substantially different cell sizes, total mRNA expression, and transcriptional activity. When existing deconvolution approaches are applied to these tissues, these systematic differences in cell sizes and transcriptomic activity confound accurate cell proportion estimates and instead may quantify total mRNA content. Furthermore, there is a lack of standard reference atlases and computational approaches to facilitate integrative analyses, including not only bulk and single cell/nuclei RNA-seq data, but also new data modalities from spatial -omic or imaging approaches. New multi-assay datasets need to be collected with orthogonal data types generated from the same tissue block and the same individual, to serve as a "gold standard" for evaluating new and existing deconvolution methods. Below, we discuss these key challenges and how they can be addressed with the acquisition of new datasets and approaches to analysis.

Evaluating performance and applications of sample-wise cell deconvolution methods on human brain transcriptomic data

Sample-wise deconvolution methods have been developed to estimate cell-type proportions and gene expressions in bulk-tissue samples. However, the performance of these methods and their biological applications has not been evaluated, particularly on human brain transcriptomic data. Here, nine deconvolution methods were evaluated with sample-matched data from bulk-tissue RNAseq, single-cell/nuclei (sc/sn) RNAseq, and immunohistochemistry. A total of 1,130,767 nuclei/cells from 149 adult postmortem brains and 72 organoid samples were used. The results showed the best performance of dtangle for estimating cell proportions and bMIND for estimating sample-wise cell-type gene expression. For eight brain cell types, 25,273 cell-type eQTLs were identified with deconvoluted expressions (decon-eQTLs). The results showed that decon-eQTLs explained more schizophrenia GWAS heritability than bulk-tissue or single-cell eQTLs alone. Differential gene expression associated with multiple phenotypes were also examined using the deconvoluted data. Our findings, which were replicated in bulk-tissue RNAseq and sc/snRNAseq data, provided new insights into the biological applications of deconvoluted data.

Detecting the effect of genetic diversity on brain composition in an Alzheimer's disease mouse model

Alzheimer's disease (AD) is characterized by neurodegeneration, pathology accumulation, and progressive cognitive decline. There is significant variation in age at onset and severity of symptoms highlighting the importance of genetic diversity in the study of AD. To address this, we analyzed cell and pathology composition of 6- and 14-month-old AD-BXD mouse brains using the semi-automated workflow (QUINT); which we expanded to allow for nonlinear refinement of brain atlas-registration, and quality control assessment of atlas-registration and brain section integrity. Near global age-related increases in microglia, astrocyte, and amyloid-beta accumulation were measured, while regional variation in neuron load existed among strains. Furthermore, hippocampal immunohistochemistry analyses were combined with bulk RNA-sequencing results to demonstrate the relationship between cell composition and gene expression. Overall, the additional functionality of the QUINT workflow delivers a highly effective method for registering and quantifying cell and pathology changes in diverse disease models.

Neuron-Glia-Ratio-Like Approach Evidenced for Limited Variability and In-Aggregate Circadian Shifts in Cortical Cell-Specific Transcriptomes

Regardless of shifts in levels of individual transcripts, it remains elusive whether natural variability in cell-specific transcriptomes within the cerebral cortex is limited in aggregate. It is also unclear whether cortical cell-specific transcriptomes might change dynamically in absence of cell number changes. Total variation in neuron- and glia-specific in-aggregate transcriptomes could be identified in a model-free way via glia-neuron ratio approach, by univariate median-to-median ratios comparing integral levels of cell-specific transcripts within a tissue sample. When deleterious, regenerative or developmental events affecting cortical cell numbers were subtle, median-to-median ratios demonstrated within-group variability not exceeding <20-25% in most cases. These levels of total variability might be explained in part by limited (~5-10%) circadian and stress-induced shifts in cell-specific cortical transcriptomes. Relevant in-aggregate transcriptomic alterations were identified after shifts in cell numbers induced by well-validated deleterious events including ischemia, traumatic injury, microglia's activation/depletion or specific mutations. Cortical median-to-median ratios also follow naturally occurring changes in the numbers of excitatory, inhibitory neurons and glial cells during perinatal brain development. These findings characterize cortical cell-specific transcriptomes as subjects to circadian shifts and lifetime events, urging the importance of reporting full details on an origin of any transcriptomic sample collected in vivo.

Identifying novel regulators of placental development using time-series transcriptome data

The placenta serves as a connection between the mother and the fetus during pregnancy, providing the fetus with oxygen, nutrients, and growth hormones. However, the regulatory mechanisms and dynamic gene interaction networks underlying early placental development are understudied. Here, we generated RNA-sequencing data from mouse fetal placenta at embryonic days 7.5, 8.5, and 9.5 to identify genes with timepoint-specific expression, then inferred gene interaction networks to analyze highly connected network modules. We determined that timepoint-specific gene network modules were associated with distinct developmental processes, and with similar expression profiles to specific human placental cell populations. From each module, we identified hub genes and their direct neighboring genes, which were predicted to govern placental functions. We confirmed that four novel candidate regulators identified through our analyses regulate cell migration in the HTR-8/SVneo cell line. Overall, we predicted several novel regulators of placental development expressed in specific placental cell types using network analysis of bulk RNA-sequencing data. Our findings and analysis approaches will be valuable for future studies investigating the transcriptional landscape of early development.

Single-cell genomics in AML: extending the frontiers of AML research

The era of genomic medicine has allowed acute myeloid leukemia (AML) researchers to improve disease characterization, optimize risk-stratification systems, and develop new treatments. Although there has been significant progress, AML remains a lethal cancer because of its remarkably complex and plastic cellular architecture. This degree of heterogeneity continues to pose a major challenge, because it limits the ability to identify and therefore eradicate the cells responsible for leukemogenesis and treatment failure. In recent years, the field of single-cell genomics has led to unprecedented strides in the ability to characterize cellular heterogeneity, and it holds promise for the study of AML. In this review, we highlight advancements in single-cell technologies, outline important shortcomings in our understanding of AML biology and clinical management, and discuss how single-cell genomics can address these shortcomings as well as provide unique opportunities in basic and translational AML research.

MiXcan: a framework for cell-type-aware transcriptome-wide association studies with an application to breast cancer

Human bulk tissue samples comprise multiple cell types with diverse roles in disease etiology. Conventional transcriptome-wide association study approaches predict genetically regulated gene expression at the tissue level, without considering cell-type heterogeneity, and test associations of predicted tissue-level expression with disease. Here we develop MiXcan, a cell-type-aware transcriptome-wide association study approach that predicts cell-type-level expression, identifies disease-associated genes via combination of cell-type-level association signals for multiple cell types, and provides insight into the disease-critical cell type. As a proof of concept, we conducted cell-type-aware analyses of breast cancer in 58,648 women and identified 12 transcriptome-wide significant genes using MiXcan compared with only eight genes using conventional approaches. Importantly, MiXcan identified genes with distinct associations in mammary epithelial versus stromal cells, including three new breast cancer susceptibility genes. These findings demonstrate that cell-type-aware transcriptome-wide analyses can reveal new insights into the genetic and cellular etiology of breast cancer and other diseases.