Gene Expression Inference Research Articles

Abstract 6258: Inference of gene expression using fragmentation patterns from targeted high-depth sequencing of cell-free DNA

Abstract Cell-free DNA (cfDNA) circulating in blood plasma arises primarily from cellular chromatin fragmentation and release due to cell death. Studies have demonstrated that the assessment of fragmentomic features of cfDNA enables gene expression inference and tissue-of-origin classification with potential applications for noninvasive cancer detection. However, due to low depth of coverage of sites of interest, current whole genome sequencing (WGS) methods have not been able to infer expression of individual genes or limited gene sets. Therefore, we developed a targeted sequencing approach that enables individual gene expression inference from cfDNA fragmentation patterns based on transcriptional start-site gene activation probability (TSS-GAP). To assess the sensitivity of this method for cfDNA-inferred gene expression profiling, we applied TSS-GAP to blends of micrococcal nuclease (MNase)-digested DNA from a cancer cell line (LS180) and sorted peripheral blood mononuclear cells (PBMC), with LS180 DNA at varying concentrations (0.1-10%). This enabled the determination of the limit of detection (LoD), relative to healthy donor background levels, on the basis of the genes’ activation probabilities (TSS-GAP scores). We found 936 genes with high inferred expression in the LS180 cancer cell line and low inferred expression in non-cancer samples (i.e. CD4, CD14, and plasma cfDNA from 4 healthy donors) and used those for healthy donor background and LoD assessment. Using these gene-level background values as cutoffs, blends were assessed to determine the lowest blend level at which expression of the gene was able to be differentiated from background. Out of the 936 genes, 326 (34.8%) were detectable at a blend level of 0.1%, 59 (6.3%) at 0.3%, 31 (3.3%) at 1%, 19 (2.0%) at 3%, and 39 (4.2%) at 10%. In total, 416 genes could be detected by TSS-GAP at levels 1% of or lower above background, highlighting the potential of our method for detecting expression of certain genes even from a small fraction of cells contributing cfDNA to liquid biopsy samples. Further expansion of this work could consider the utility of such a method for circulating tumor DNA (ctDNA) detection, cfDNA tissue-of-origin classification, and deconvolution of cell types contributing to cfDNA found in blood. Citation Format: Emily Leff, Alexander Tseng, Victoria Cheung, Ehsan Tabari, Kimberly Walter, Alex Lovejoy, Cheng-Ho J. Lin. Inference of gene expression using fragmentation patterns from targeted high-depth sequencing of cell-free DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6258.

Chromatin- and nucleosome-associated features in liquid biopsy: implications for cancer biomarker discovery.

Cell-free DNA (cfDNA) from the bloodstream has been studied for cancer biomarker discovery, and chromatin-derived epigenetic features have come into the spotlight for their potential to expand clinical applications. Methylation, fragmentation, and nucleosome positioning patterns of cfDNA have previously been shown to reveal epigenomic and inferred transcriptomic information. More recently, histone modifications have emerged as a tool to further identify tumor-specific chromatin variants in plasma. A number of sequencing methods have been developed to analyze these epigenetic markers, offering new insights into tumor biology. Features within cfDNA allow for cancer detection, subtype and tissue of origin classification, and inference of gene expression. These methods provide a window into the complexity of cancer and the dynamic nature of its progression. In this review, we highlight the array of epigenetic features in cfDNA that can be extracted from chromatin- and nucleosome-associated organization and outline potential use cases in cancer management.

Longitudinal Noninvasive Surveillance & Fragmentomic Characterization of Follicular Lymphoma

Introduction While cell-free DNA (cfDNA) plays an increasingly defined role in aggressive lymphomas, its characteristics in indolent lymphomas are less established. Many follicular lymphoma (FL) patients experience durable remissions and late relapses ( Fig A). We therefore explored circulating tumor DNA (ctDNA) kinetics during long-term blood-based surveillance in patients with FL. In addition to noninvasive detection using somatic mutations, we also evaluated gene expression inference from cfDNA utilizing a novel fragmentomic method to detect histological transformation (tFL). Methods We studied 121 serial samples in a cohort of 17 FL patients (median 7 per patient) with long term clinical and blood-based MRD monitoring (median 5.6 years, max 10.8). We considered serial ctDNA status at 4 milestones: baseline/pre-treatment, post-treatment, in radiographic remission (MRD), and at relapse. We also profiled pre-treatment cfDNA for 9 tFL cases. Cases were genotyped to identify somatic mutations from diagnostic tumor or baseline plasma, with matched PBMCs used to censor germline variants and clonal hematopoiesis. ctDNA status in subsequent samples was evaluated via CAPP-Seq (SNVs) and PhasED-Seq (PVs). The cfDNA fragmentation profiles of pre-treatment FL and control samples were evaluated for inferred gene expression via EPIC-seq (Esfahani Nat Biotech 2022) using a novel 1676-gene panel focused on lymphoid neoplasms, including classic and transformed FL. Results Treatment & Relapse Kinetics Most patients had advanced disease (88% stage III-IV) and Low to Intermediate risk FLIPI (71% 0-2). At first cfDNA evaluation, 9/17 (53%) were previously untreated. Median pre-treatment ctDNA levels were 73.4 hGE/mL in FL, as compared to 236 hGE/mL in DLBCL (Kurtz JCO 2018). Post-treatment, median ctDNA levels dropped to 1.2 hGE/mL. Depth of response was similar between rituximab monotherapy and cytotoxic regimens, with median -2.2 log10 fold change from baseline. MRD Detection PVs were observed in all tumors, consistent with expected aberrant somatic hypermutation in FL. To evaluate MRD detection via PhasED-Seq, we considered plasma timepoints with undetectable ctDNA by CAPP-Seq (n=26) from patients (n=9) with subsequent clinical relapse. MRD was detected in 11 additional specimens (42%) by PhasED-Seq. Among FL patients tested for MRD while in radiographic remission, low-level ctDNA was detectable in 62% of plasma specimens (8/13) obtained within 1 month of negative PET/CT, suggesting potential additional utility for surveillance. Long-term Genomic Stability Sequential biopsies of FL tumors have shown genomic heterogeneity between diagnosis and relapse. We therefore evaluated ctDNA concordance across variant classes & regions. Relative AF was calculated pre-treatment and at relapse (median 2 years between samples, range 1-8 years), normalized by total ctDNA levels. The most stable lesions included missense mutations in CREBBP and TP53, nonsense KMT2D mutations, and BCL2 5'-UTR variants. BCL6 intronic mutations and chr2/chr14 SHM varied between diagnosis/relapse. Plasma versus Cellular Concentrations In contrast to DLBCL, circulating FL cells are frequently detectable at low levels in blood. To compare relative FL concentrations in the cellular versus plasma compartments, we profiled 10 paired specimens from patients with radiographic disease. While tumor-confirmed variants were detected in 7 of 10 cellular specimens, mean allelic frequency (AF) was 5.3x higher in plasma. Fragmentomic Assessment of Transformation We compared gene expression profiles inferred from cfDNA from pre-treatment lymphoma (FL=15, tFL=9) versus control (n=20) samples ( Fig B). Considering genes associated with FL (Huet Lancet Onc 2018), FL patients had significantly higher inferred expression than healthy adults (p=0.036). Likewise, patients with tFL showed higher inferred expression of genes associated with histologic transformation (Gentles Blood 2009, p=0.025). Conclusions Sensitive methods are required for effective MRD assessment in FL given lower ctDNA shedding as compared with DLBCL. Frequent detectable MRD in radiographic remission suggests that integrating molecular surveillance may augment serial imaging in the long-term management of FL. Simultaneous inferred gene expression from FL cfDNA appears to hold promise, including for noninvasive detection of transformation.

LipGene: Lipschitz Continuity Guided Adaptive Learning Rates for Fast Convergence on Microarray Expression Data Sets.

Hyperparameter tuning, specifically tuning of learning rate, can often be a time-consuming process, especially when dealing with large data sets. A mathematical foundation in the choice of learning rate can minimize tuning efforts. We propose the application of a novel adaptive learning rate paradigm, guided by Lipschitz continuity of the loss functions (LipGene), to the task of Gene Expression Inference using shallow neural networks. We utilize Mean Absolute Error and Quantile loss separately for training. Our adaptive learning rate, which is dynamically computed for each epoch, is based on the principle of Lipschitz constant and requires no tuning. Experimentally, we prove that our proposed approach greatly surpasses conventional choices of learning rates in terms of both speed of convergence and generalizability. Advocating the principle of Parsimonious Computing, our method can reduce compute infrastructure required for training by using smaller networks with a minimal compromise on the prediction error.

Inference and uncertainty quantification of stochastic gene expression via synthetic models.

Estimating uncertainty in model predictions is a central task in quantitative biology. Biological models at the single-cell level are intrinsically stochastic and nonlinear, creating formidable challenges for their statistical estimation which inevitably has to rely on approximations that trade accuracy for tractability. Despite intensive interest, a sweet spot in this trade-off has not been found yet. We propose a flexible procedure for uncertainty quantification in a wide class of reaction networks describing stochastic gene expression including those with feedback. The method is based on creating a tractable coarse-graining of the model that is learned from simulations, a synthetic model, to approximate the likelihood function. We demonstrate that synthetic models can substantially outperform state-of-the-art approaches on a number of non-trivial systems and datasets, yielding an accurate and computationally viable solution to uncertainty quantification in stochastic models of gene expression.

Abstract 5031: Improved deconvolution of combined bulk and single-cell RNA-sequencing data

Abstract Identifying how cell types and their abundances evolve during tumor progression is critical to understanding the mechanisms and identifying predictors of metastasis. Single-cell RNA sequencing (scRNA-seq) has been especially promising in resolving heterogeneity of expression programs at the single cell level but is not always available, for example for large cohort studies or longitudinal analysis of archived samples. In such cases, cell subpopulations must be inferred by deconvolution, a process that can infer single-cell genomic data from bulk data but has limited ability to resolve fine clonal structure. We extend our previous bulk genomic deconvolution tool, Robust and Accurate Deconvolution (RAD), to establish a new method, scRAD, that can use reference scRNA-seq to interpret sample collections for which only bulk RNA-seq is available for some samples, e.g., clonally resolving archived primary (PRM) tissues using scRNA-seq from metastases (METs). We preprocess scRNA-seq data to accurately represent gene expression profiles, yielding a signature matrix S then extend our RAD method via a regularization term to deconvolve bulk data while maximizing consistency with S. We validate our method on semi-synthetic data derived from human PRM breast cancer cases and bone and ovary METs, showing that scRAD improves inference of single-cell gene expression profiles and their frequencies relative to the prior RAD with random initialization or initialization using the single-cell matrix S (Table 1). We then apply scRAD to a collection of paired PRM and MET tumors to quantify progression changes in common cell types. One-sided Kaplan-Meier analysis shows that tumors inferred to increase the mast cell fraction from PRM to MET exhibit lower overall survival (p<0.05), consistent with the role of mast cells in metastatic growth and propagation. Tumors that show increased macrophage cell fraction from PRM to MET show improved overall survival (p<0.04), consistent with the role of immune infiltration in survival. mean square error (MSE) of gene expression and mixture fraction inference on semi-simulated data Method RAD with random initialization RAD with random initialization RAD with random initialization RAD initialized with S RAD initialized with S RAD initialized with S scRAD scRAD scRAD Sample number 2 4 8 2 4 8 2 4 8 Gene Expression MSE 0.41 0.28 0.37 0.31 0.27 0.37 0.22 0.15 0.13 Mixture fraction MSE 0.58 0.82 0.59 0.39 0.83 0.58 0.19 0.20 0.22 Citation Format: Haoyun Lei, Xiaoyan A. Guo, Yifeng Tao, Kai Ding, Xuecong Fu, Steffi Oesterreich, Adrian V. Lee, Russell Schwartz. Improved deconvolution of combined bulk and single-cell RNA-sequencing data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 5031.

Deep Large-Scale Multitask Learning Network for Gene Expression Inference.

Gene expression profiling makes it possible to conduct many biological studies in a variety of fields due to its thorough characterization of cellular states under various experimental conditions. Despite recent advances in high-throughput technology, profiling an entire set of genomes is still difficult and expensive. Due to the high correlation between expression patterns of different genes, the aforementioned problem can be solved with a cost-effective approach that collects only a small subset of genes, called landmark genes, representing the entire set of genes, and infer the remaining genes, called target genes, using a computational model. There are several shallow and deep regression models in literature to estimate the expressions of target genes from the landmark genes. However, the shallow mostly have limited capacity in learning the nonlinear and complex gene expression data and are prone to underfitting, and the deep models generally do not take advantage of correlation among target genes in the learning process and suffer from overfitting. Considering the gene expression inference as a multitask learning problem, we propose a new deep multitask learning algorithm to tackle these issues. Our learning framework automatically learns the correlation between target genes and uses this knowledge to improve its generalization. Specifically, we utilize a subnetwork with low-dimensional latent variables to discover the relationships between target genes and enforce a seamless and easy to implement regularization to our deep regression model. Unlike the existing multitask learning methods that can only deal with dozens or hundreds of tasks, our algorithm is able to efficiently learn the relationships between ∼10,000 target genes and, thus, is scalable to a large number of tasks. Our proposed method outperforms the shallow and deep regression models for gene expression inference and alternative multitask learning algorithms on two large-scale datasets regardless of the network architecture.

PseudotimeDE: inference of differential gene expression along cell pseudotime with well-calibrated p-values from single-cell RNA sequencing data

To investigate molecular mechanisms underlying cell state changes, a crucial analysis is to identify differentially expressed (DE) genes along the pseudotime inferred from single-cell RNA-sequencing data. However, existing methods do not account for pseudotime inference uncertainty, and they have either ill-posed p-values or restrictive models. Here we propose PseudotimeDE, a DE gene identification method that adapts to various pseudotime inference methods, accounts for pseudotime inference uncertainty, and outputs well-calibrated p-values. Comprehensive simulations and real-data applications verify that PseudotimeDE outperforms existing methods in false discovery rate control and power.

PseudotimeDE: inference of differential gene expression along cell pseudotime with well-calibrated p-values from single-cell RNA sequencing data

PseudotimeDE: inference of differential gene expression along cell pseudotime with well-calibrated p-values from single-cell RNA sequencing data

On transformative adaptive activation functions in neural networks for gene expression inference.

Gene expression profiling was made more cost-effective by the NIH LINCS program that profiles only ∼1, 000 selected landmark genes and uses them to reconstruct the whole profile. The D-GEX method employs neural networks to infer the entire profile. However, the original D-GEX can be significantly improved. We propose a novel transformative adaptive activation function that improves the gene expression inference even further and which generalizes several existing adaptive activation functions. Our improved neural network achieves an average mean absolute error of 0.1340, which is a significant improvement over our reimplementation of the original D-GEX, which achieves an average mean absolute error of 0.1637. The proposed transformative adaptive function enables a significantly more accurate reconstruction of the full gene expression profiles with only a small increase in the complexity of the model and its training procedure compared to other methods.

On tower and checkerboard neural network architectures for gene expression inference

BackgroundOne possible approach how to economically facilitate gene expression profiling is to use the L1000 platform which measures the expression of ∼1,000 landmark genes and uses a computational method to infer the expression of another ∼10,000 genes. One such method for the gene expression inference is a D–GEX which employs neural networks.ResultsWe propose two novel D–GEX architectures that significantly improve the quality of the inference by increasing the capacity of a network without any increase in the number of trained parameters. The architectures partition the network into individual towers. Our best proposed architecture — a checkerboard architecture with a skip connection and five towers — together with minor changes in the training protocol improves the average mean absolute error of the inference from 0.134 to 0.128.ConclusionsOur proposed approach increases the gene expression inference accuracy without increasing the number of weights of the model and thus without increasing the memory footprint of the model that is limiting its usage.

DISC: a highly scalable and accurate inference of gene expression and structure for single-cell transcriptomes using semi-supervised deep learning

Dropouts distort gene expression and misclassify cell types in single-cell transcriptome. Although imputation may improve gene expression and downstream analysis to some degree, it also inevitably introduces false signals. We develop DISC, a novel deep learning network with semi-supervised learning to infer gene structure and expression obscured by dropouts. Compared with seven state-of-the-art imputation approaches on ten real-world datasets, we show that DISC consistently outperforms the other approaches. Its applicability, scalability, and reliability make DISC a promising approach to recover gene expression, enhance gene and cell structures, and improve cell type identification for sparse scRNA-seq data.

DISC: a highly scalable and accurate inference of gene expression and structure for single-cell transcriptomes using semi-supervised deep learning.

DISC: a highly scalable and accurate inference of gene expression and structure for single-cell transcriptomes using semi-supervised deep learning.

Blood-based multi-tissue gene expression inference with Bayesian ridge regression

Gene expression profiling is widely used in basic and cancer research but still not feasible in many clinical applications because tissues, such as brain samples, are difficult and not ethnical to collect. Gene expression in uncollected tissues can be computationally inferred using genotype and expression quantitative trait loci. No methods can infer unmeasured gene expression of multiple tissues with single tissue gene expression profile as input. Here, we present a Bayesian ridge regression-based method (B-GEX) to infer gene expression profiles of multiple tissues from blood gene expression profile. For each gene in a tissue, a low-dimensional feature vector was extracted from whole blood gene expression profile by feature selection. We used GTEx RNAseq data of 16 tissues to train inference models to capture the cross-tissue expression correlations between each target gene in a tissue and its preselected feature genes in peripheral blood. We compared B-GEX with least square regression, LASSO regression and ridge regression. B-GEX outperforms the other three models in most tissues in terms of mean absolute error, Pearson correlation coefficient and root-mean-squared error. Moreover, B-GEX infers expression level of tissue-specific genes as well as those of non-tissue-specific genes in all tissues. Unlike previous methods, which require genomic features or gene expression profiles of multiple tissues, our model only requires whole blood expression profile as input. B-GEX helps gain insights into gene expressions of uncollected tissues from more accessible data of blood. B-GEX is available at https://github.com/xuwenjian85/B-GEX. Supplementary data are available at Bioinformatics online.

Stochastic reaction networks with input processes: Analysis and application to gene expression inference

Stochastic reaction network modelling is widely utilized to describe the probabilistic dynamics of biochemical systems in general, and gene interaction networks in particular. The statistical analysis of the response of these systems to perturbation inputs is typically dependent on specific perturbation models. Motivated by reporter gene systems, widely utilized in biology to monitor gene activity in individual cells, we address the analysis of reaction networks with state-affine rates in presence of an input process. We develop a generalization of the so-called moment equations that precisely accounts for the first- and second-order moments of arbitrary inputs without the need for a model of the input process, as well as spectral relationships between the network input and state. We then apply these results to develop a method for the reconstruction of the autocovariance function of gene activity from reporter gene population-snapshot data, a crucial step toward the investigation of gene regulation, and demonstrate its performance on a simulated case study.

Reproducibility of Molecular Phenotypes after Long-Term Differentiation toHuman iPSC-Derived Neurons: A Multi-Site Omics Study.

SummaryReproducibility in molecular and cellular studies is fundamental to scientific discovery. To establish the reproducibility of a well-defined long-term neuronal differentiation protocol, we repeated the cellular and molecular comparison of the same two iPSC lines across five distinct laboratories. Despite uncovering acceptable variability within individual laboratories, we detect poor cross-site reproducibility of the differential gene expression signature between these two lines. Factor analysis identifies the laboratory as the largest source of variation along with several variation-inflating confounders such as passaging effects and progenitor storage. Single-cell transcriptomics shows substantial cellular heterogeneity underlying inter-laboratory variability and being responsible for biases in differential gene expression inference. Factor analysis-based normalization of the combined dataset can remove the nuisance technical effects, enabling the execution of robust hypothesis-generating studies. Our study shows that multi-center collaborations can expose systematic biases and identify critical factors to be standardized when publishing novel protocols, contributing to increased cross-site reproducibility.

Conditional generative adversarial network for gene expression inference.

The rapid progress of gene expression profiling has facilitated the prosperity of recent biological studies in various fields, where gene expression data characterizes various cell conditions and regulatory mechanisms under different experimental circumstances. Despite the widespread application of gene expression profiling and advances in high-throughput technologies, profiling in genome-wide level is still expensive and difficult. Previous studies found that high correlation exists in the expression pattern of different genes, such that a small subset of genes can be informative to approximately describe the entire transcriptome. In the Library of Integrated Network-based Cell-Signature program, a set of ∼1000 landmark genes have been identified that contain ∼80% information of the whole genome and can be used to predict the expression of remaining genes. For a cost-effective profiling strategy, traditional methods measure the profiles of landmark genes and then infer the expression of other target genes via linear models. However, linear models do not have the capacity to capture the non-linear associations in gene regulatory networks. As a flexible model with high representative power, deep learning models provide an alternate to interpret the complex relation among genes. In this paper, we propose a deep learning architecture for the inference of target gene expression profiles. We construct a novel conditional generative adversarial network by incorporating both the adversarial and ℓ1-norm loss terms in our model. Unlike the smooth and blurry predictions resulted by mean squared error objective, the coupled adversarial and ℓ1-norm loss function leads to more accurate and sharp predictions. We validate our method under two different settings and find consistent and significant improvements over all the comparing methods.

Bayesian Inference of Allele-Specific Gene Expression Indicates Abundant Cis-Regulatory Variation in Natural Flycatcher Populations.

Polymorphism in cis-regulatory sequences can lead to different levels of expression for the two alleles of a gene, providing a starting point for the evolution of gene expression. Little is known about the genome-wide abundance of genetic variation in gene regulation in natural populations but analysis of allele-specific expression (ASE) provides a means for investigating such variation. We performed RNA-seq of multiple tissues from population samples of two closely related flycatcher species and developed a Bayesian algorithm that maximizes data usage by borrowing information from the whole data set and combines several SNPs per transcript to detect ASE. Of 2,576 transcripts analyzed in collared flycatcher, ASE was detected in 185 (7.2%) and a similar frequency was seen in the pied flycatcher. Transcripts with statistically significant ASE commonly showed the major allele in >90% of the reads, reflecting that power was highest when expression was heavily biased toward one of the alleles. This would suggest that the observed frequencies of ASE likely are underestimates. The proportion of ASE transcripts varied among tissues, being lowest in testis and highest in muscle. Individuals often showed ASE of particular transcripts in more than one tissue (73.4%), consistent with a genetic basis for regulation of gene expression. The results suggest that genetic variation in regulatory sequences commonly affects gene expression in natural populations and that it provides a seedbed for phenotypic evolution via divergence in gene expression.

Imputing gene expression to maximize platform compatibility.

Microarray measurements of gene expression constitute a large fraction of publicly shared biological data, and are available in the Gene Expression Omnibus (GEO). Many studies use GEO data to shape hypotheses and improve statistical power. Within GEO, the Affymetrix HG-U133A and HG-U133 Plus 2.0 are the two most commonly used microarray platforms for human samples; the HG-U133 Plus 2.0 platform contains 54 220 probes and the HG-U133A array contains a proper subset (21 722 probes). When different platforms are involved, the subset of common genes is most easily compared. This approach results in the exclusion of substantial measured data and can limit downstream analysis. To predict the expression values for the genes unique to the HG-U133 Plus 2.0 platform, we constructed a series of gene expression inference models based on genes common to both platforms. Our model predicts gene expression values that are within the variability observed in controlled replicate studies and are highly correlated with measured data. Using six previously published studies, we also demonstrate the improved performance of the enlarged feature space generated by our model in downstream analysis.Availability and ImplementationThe gene inference model described in this paper is available as a R package (affyImpute), which can be downloaded at http://simtk.org/home/affyimpute.Supplementary information Supplementary data are available at Bioinformatics online.

CellMapper: rapid and accurate inference of gene expression in difficult-to-isolate cell types.

We present a sensitive approach to predict genes expressed selectively in specific cell types, by searching publicly available expression data for genes with a similar expression profile to known cell-specific markers. Our method, CellMapper, strongly outperforms previous computational algorithms to predict cell type-specific expression, especially for rare and difficult-to-isolate cell types. Furthermore, CellMapper makes accurate predictions for human brain cell types that have never been isolated, and can be rapidly applied to diverse cell types from many tissues. We demonstrate a clinically relevant application to prioritize candidate genes in disease susceptibility loci identified by GWAS.Electronic supplementary materialThe online version of this article (doi:10.1186/s13059-016-1062-5) contains supplementary material, which is available to authorized users.