Compressing gene expression data using multiple latent space dimensionalities learns complementary biological representations

A deep profile of gene expression across 18 human cancers

Clinically and biologically valuable information may reside untapped in large cancer gene expression data sets. Deep unsupervised learning has the potential to extract this information with unprecedented efficacy but has thus far been hampered by a lack of biological interpretability and robustness. Here, we present DeepProfile, a comprehensive framework that addresses current challenges in applying unsupervised deep learning to gene expression profiles. We use DeepProfile to learn low-dimensional latent spaces for 18 human cancers from 50,211 transcriptomes. DeepProfile outperforms existing dimensionality reduction methods with respect to biological interpretability. Using DeepProfile interpretability methods, we show that genes that are universally important in defining the latent spaces across all cancer types control immune cell activation, while cancer type-specific genes and pathways define molecular disease subtypes. By linking DeepProfile latent variables to secondary tumor characteristics, we discover that tumor mutation burden is closely associated with the expression of cell cycle-related genes. DNA mismatch repair and MHC class II antigen presentation pathway expression, on the other hand, are consistently associated with patient survival. We validate these results through Kaplan-Meier analyses and nominate tumor-associated macrophages as an important source of survival-correlated MHC class II transcripts. Our results illustrate the power of unsupervised deep learning for discovery of cancer biology from existing gene expression data.

iModulonMiner and PyModulon: Software for unsupervised mining of gene expression compendia

Public gene expression databases are a rapidly expanding resource of organism responses to diverse perturbations, presenting both an opportunity and a challenge for bioinformatics workflows to extract actionable knowledge of transcription regulatory network function. Here, we introduce a five-step computational pipeline, called iModulonMiner, to compile, process, curate, analyze, and characterize the totality of RNA-seq data for a given organism or cell type. This workflow is centered around the data-driven computation of co-regulated gene sets using Independent Component Analysis, called iModulons, which have been shown to have broad applications. As a demonstration, we applied this workflow to generate the iModulon structure of Bacillus subtilis using all high-quality, publicly-available RNA-seq data. Using this structure, we predicted regulatory interactions for multiple transcription factors, identified groups of co-expressed genes that are putatively regulated by undiscovered transcription factors, and predicted properties of a recently discovered single-subunit phage RNA polymerase. We also present a Python package, PyModulon, with functions to characterize, visualize, and explore computed iModulons. The pipeline, available at https://github.com/SBRG/iModulonMiner, can be readily applied to diverse organisms to gain a rapid understanding of their transcriptional regulatory network structure and condition-specific activity.

BuDDI: BulkDeconvolution withDomainInvariance to predict cell-type-specific perturbations from bulk

While single-cell experiments provide deep cellular resolution within a single sample, some single-cell experiments are inherently more challenging than bulk experiments due to dissociation difficulties, cost, or limited tissue availability. This creates a situation where we have deep cellular profiles of one sample or condition, and bulk profiles across multiple samples and conditions. To bridge this gap, we propose BuDDI (BUlk Deconvolution with Domain Invariance). BuDDI utilizes domain adaptation techniques to effectively integrate available corpora of case-control bulk and reference scRNA-seq observations to infer cell-type-specific perturbation effects. BuDDI achieves this by learning independent latent spaces within a single variational autoencoder (VAE) encompassing at least four sources of variability: 1) cell type proportion, 2) perturbation effect, 3) structured experimental variability, and 4) remaining variability. Since each latent space is encouraged to be independent, we simulate perturbation responses by independently composing each latent space to simulate cell-type-specific perturbation responses. We evaluated BuDDI's performance on simulated and real data with experimental designs of increasing complexity. We first validated that BuDDI could learn domain invariant latent spaces on data with matched samples across each source of variability. Then we validated that BuDDI could accurately predict cell-type-specific perturbation response when no single-cell perturbed profiles were used during training; instead, only bulk samples had both perturbed and non-perturbed observations. Finally, we validated BuDDI on predicting sex-specific differences, an experimental design where it is not possible to have matched samples. In each experiment, BuDDI outperformed all other comparative methods and baselines. As more reference atlases are completed, BuDDI provides a path to combine these resources with bulk-profiled treatment or disease signatures to study perturbations, sex differences, or other factors at single-cell resolution.

Optimizer's dilemma: optimization strongly influences model selection in transcriptomic prediction

Motivation

Most models can be fit to data using various optimization approaches. While model choice is frequently reported in machine-learning-based research, optimizers are not often noted. We applied two different implementations of LASSO logistic regression implemented in Python's scikit-learn package, using two different optimization approaches (coordinate descent, implemented in the liblinear library, and stochastic gradient descent, or SGD), to predict mutation status and gene essentiality from gene expression across a variety of pan-cancer driver genes. For varying levels of regularization, we compared performance and model sparsity between optimizers.

Results

After model selection and tuning, we found that liblinear and SGD tended to perform comparably. liblinear models required more extensive tuning of regularization strength, performing best for high model sparsities (more nonzero coefficients), but did not require selection of a learning rate parameter. SGD models required tuning of the learning rate to perform well, but generally performed more robustly across different model sparsities as regularization strength decreased. Given these tradeoffs, we believe that the choice of optimizers should be clearly reported as a part of the model selection and validation process, to allow readers and reviewers to better understand the context in which results have been generated.

Availability and implementation

The code used to carry out the analyses in this study is available at https://github.com/greenelab/pancancer-evaluation/tree/master/01_stratified_classification. Performance/regularization strength curves for all genes in the Vogelstein et al. (2013) dataset are available at https://doi.org/10.6084/m9.figshare.22728644.

Immune, metabolic landscapes of prognostic signatures for lung adenocarcinoma based on a novel deep learning framework

Lung adenocarcinoma (LUAD) is a malignant tumor with high lethality, and the aim of this study was to identify promising biomarkers for LUAD. Using the TCGA-LUAD dataset as a discovery cohort, a novel joint framework VAEjMLP based on variational autoencoder (VAE) and multilayer perceptron (MLP) was proposed. And the Shapley Additive Explanations (SHAP) method was introduced to evaluate the contribution of feature genes to the classification decision, which helped us to develop a biologically meaningful biomarker potential scoring algorithm. Nineteen potential biomarkers for LUAD were identified, which were involved in the regulation of immune and metabolic functions in LUAD. A prognostic risk model for LUAD was constructed by the biomarkers HLA-DRB1, SCGB1A1, and HLA-DRB5 screened by Cox regression analysis, dividing the patients into high-risk and low-risk groups. The prognostic risk model was validated with external datasets. The low-risk group was characterized by enrichment of immune pathways and higher immune infiltration compared to the high-risk group. While, the high-risk group was accompanied by an increase in metabolic pathway activity. There were significant differences between the high- and low-risk groups in metabolic reprogramming of aerobic glycolysis, amino acids, and lipids, as well as in angiogenic activity, epithelial-mesenchymal transition, tumorigenic cytokines, and inflammatory response. Furthermore, high-risk patients were more sensitive to Afatinib, Gefitinib, and Gemcitabine as predicted by the pRRophetic algorithm. This study provides prognostic signatures capable of revealing the immune and metabolic landscapes for LUAD, and may shed light on the identification of other cancer biomarkers.

Inferring cellular and molecular processes in single-cell data with non-negative matrix factorization using Python, R and GenePattern Notebook implementations of CoGAPS

Non-negative matrix factorization (NMF) is an unsupervised learning method well suited to high-throughput biology. However, inferring biological processes from an NMF result still requires additional post hoc statistics and annotation for interpretation of learned features. Here, we introduce a suite of computational tools that implement NMF and provide methods for accurate and clear biological interpretation and analysis. A generalized discussion of NMF covering its benefits, limitations and open questions is followed by four procedures for the Bayesian NMF algorithm Coordinated Gene Activity across Pattern Subsets (CoGAPS). Each procedure will demonstrate NMF analysis to quantify cell state transitions in a public domain single-cell RNA-sequencing dataset. The first demonstrates PyCoGAPS, our new Python implementation that enhances runtime for large datasets, and the second allows its deployment in Docker. The third procedure steps through the same single-cell NMF analysis using our R CoGAPS interface. The fourth introduces a beginner-friendly CoGAPS platform using GenePattern Notebook, aimed at users with a working conceptual knowledge of data analysis but without a basic proficiency in the R or Python programming language. We also constructed a user-facing website to serve as a central repository for information and instructional materials about CoGAPS and its application programming interfaces. The expected timing to setup the packages and conduct a test run is around 15 min, and an additional 30 min to conduct analyses on a precomputed result. The expected runtime on the user's desired dataset can vary from hours to days depending on factors such as dataset size or input parameters.

Machine learning in rare disease

High-throughput profiling methods (such as genomics or imaging) have accelerated basic research and made deep molecular characterization of patient samples routine. These approaches provide a rich portrait of genes, molecular pathways and cell types involved in disease phenotypes. Machine learning (ML) can be a useful tool for extracting disease-relevant patterns from high-dimensional datasets. However, depending upon the complexity of the biological question, machine learning often requires many samples to identify recurrent and biologically meaningful patterns. Rare diseases are inherently limited in clinical cases, leading to few samples to study. In this Perspective, we outline the challenges and emerging solutions for using ML for small sample sets, specifically in rare diseases. Advances in ML methods for rare diseases are likely to be informative for applications beyond rare diseases for which few samples exist with high-dimensional data. We propose that the method community prioritize the development of ML techniques for rare disease research.

Deep Transcriptome Profiling of Multiple Myeloma Using Quantitative Phenotypes

BACKGROUND

Transcriptome studies are gaining momentum in genomic epidemiology, and the need to incorporate these data in multivariable models alongside other risk factors brings demands for new approaches.

METHODS

Here we describe SPECTRA, an approach to derive quantitative variables that capture the intrinsic variation in gene expression of a tissue type. We applied the SPECTRA approach to bulk RNA sequencing from malignant cells (CD138+) in patients from the Multiple Myeloma Research Foundation CoMMpass study.

RESULTS

A set of 39 spectra variables were derived to represent multiple myeloma cells. We used these variables in predictive modeling to determine spectra-based risk scores for overall survival, progression-free survival, and time to treatment failure. Risk scores added predictive value beyond known clinical and expression risk factors and replicated in an external dataset. Spectrum variable S5, a significant predictor for all three outcomes, showed pre-ranked gene set enrichment for the unfolded protein response, a mechanism targeted by proteasome inhibitors which are a common first line agent in multiple myeloma treatment. We further used the 39 spectra variables in descriptive modeling, with significant associations found with tumor cytogenetics, race, gender, and age at diagnosis; factors known to influence multiple myeloma incidence or progression.

CONCLUSIONS

Quantitative variables from the SPECTRA approach can predict clinical outcomes in multiple myeloma and provide a new avenue for insight into tumor differences by demographic groups.

IMPACT

The SPECTRA approach provides a set of quantitative phenotypes that deeply profile a tissue and allows for more comprehensive modeling of gene expression with other risk factors.

PAUSE: principled feature attribution for unsupervised gene expression analysis

As interest in using unsupervised deep learning models to analyze gene expression data has grown, an increasing number of methods have been developed to make these models more interpretable. These methods can be separated into two groups: post hoc analyses of black box models through feature attribution methods and approaches to build inherently interpretable models through biologically-constrained architectures. We argue that these approaches are not mutually exclusive, but can in fact be usefully combined. We propose PAUSE ( https://github.com/suinleelab/PAUSE ), an unsupervised pathway attribution method that identifies major sources of transcriptomic variation when combined with biologically-constrained neural network models.

Predicting compound activity from phenotypic profiles and chemical structures

Predicting assay results for compounds virtually using chemical structures and phenotypic profiles has the potential to reduce the time and resources of screens for drug discovery. Here, we evaluate the relative strength of three high-throughput data sources-chemical structures, imaging (Cell Painting), and gene-expression profiles (L1000)-to predict compound bioactivity using a historical collection of 16,170 compounds tested in 270 assays for a total of 585,439 readouts. All three data modalities can predict compound activity for 6-10% of assays, and in combination they predict 21% of assays with high accuracy, which is a 2 to 3 times higher success rate than using a single modality alone. In practice, the accuracy of predictors could be lower and still be useful, increasing the assays that can be predicted from 37% with chemical structures alone up to 64% when combined with phenotypic data. Our study shows that unbiased phenotypic profiling can be leveraged to enhance compound bioactivity prediction to accelerate the early stages of the drug-discovery process.

Cross-platform normalization enables machine learning model training on microarray and RNA-seq data simultaneously

Large compendia of gene expression data have proven valuable for the discovery of novel biological relationships. Historically, most available RNA assays were run on microarray, while RNA-seq is now the platform of choice for many new experiments. The data structure and distributions between the platforms differ, making it challenging to combine them directly. Here we perform supervised and unsupervised machine learning evaluations to assess which existing normalization methods are best suited for combining microarray and RNA-seq data. We find that quantile and Training Distribution Matching normalization allow for supervised and unsupervised model training on microarray and RNA-seq data simultaneously. Nonparanormal normalization and z-scores are also appropriate for some applications, including pathway analysis with Pathway-Level Information Extractor (PLIER). We demonstrate that it is possible to perform effective cross-platform normalization using existing methods to combine microarray and RNA-seq data for machine learning applications.

resVAE ensemble: Unsupervised identification of gene sets in multi-modal single-cell sequencing data using deep ensembles

Feature identification and manual inspection is currently still an integral part of biological data analysis in single-cell sequencing. Features such as expressed genes and open chromatin status are selectively studied in specific contexts, cell states or experimental conditions. While conventional analysis methods construct a relatively static view on gene candidates, artificial neural networks have been used to model their interactions after hierarchical gene regulatory networks. However, it is challenging to identify consistent features in this modeling process due to the inherently stochastic nature of these methods. Therefore, we propose using ensembles of autoencoders and subsequent rank aggregation to extract consensus features in a less biased manner. Here, we performed sequencing data analyses of different modalities either independently or simultaneously as well as with other analysis tools. Our resVAE ensemble method can successfully complement and find additional unbiased biological insights with minimal data processing or feature selection steps while giving a measurement of confidence, especially for models using stochastic or approximation algorithms. In addition, our method can also work with overlapping clustering identity assignment suitable for transitionary cell types or cell fates in comparison to most conventional tools.

Causal Discovery in High-dimensional, Multicollinear Datasets

As the cost of high-throughput genomic sequencing technology declines, its application in clinical research becomes increasingly popular. The collected datasets often contain tens or hundreds of thousands of biological features that need to be mined to extract meaningful information. One area of particular interest is discovering underlying causal mechanisms of disease outcomes. Over the past few decades, causal discovery algorithms have been developed and expanded to infer such relationships. However, these algorithms suffer from the curse of dimensionality and multicollinearity. A recently introduced, non-orthogonal, general empirical Bayes approach to matrix factorization has been demonstrated to successfully infer latent factors with interpretable structures from observed variables. We hypothesize that applying this strategy to causal discovery algorithms can solve both the high dimensionality and collinearity problems, inherent to most biomedical datasets. We evaluate this strategy on simulated data and apply it to two real-world datasets. In a breast cancer dataset, we identified important survival-associated latent factors and biologically meaningful enriched pathways within factors related to important clinical features. In a SARS-CoV-2 dataset, we were able to predict whether a patient (1) had Covid-19 and (2) would enter the ICU. Furthermore, we were able to associate factors with known Covid-19 related biological pathways.

Using genome-wide expression compendia to study microorganisms

A gene expression compendium is a heterogeneous collection of gene expression experiments assembled from data collected for diverse purposes. The widely varied experimental conditions and genetic backgrounds across samples creates a tremendous opportunity for gaining a systems level understanding of the transcriptional responses that influence phenotypes. Variety in experimental design is particularly important for studying microbes, where the transcriptional responses integrate many signals and demonstrate plasticity across strains including response to what nutrients are available and what microbes are present. Advances in high-throughput measurement technology have made it feasible to construct compendia for many microbes. In this review we discuss how these compendia are constructed and analyzed to reveal transcriptional patterns.

Variational autoencoders learn transferrable representations of metabolomics data

Dimensionality reduction approaches are commonly used for the deconvolution of high-dimensional metabolomics datasets into underlying core metabolic processes. However, current state-of-the-art methods are widely incapable of detecting nonlinearities in metabolomics data. Variational Autoencoders (VAEs) are a deep learning method designed to learn nonlinear latent representations which generalize to unseen data. Here, we trained a VAE on a large-scale metabolomics population cohort of human blood samples consisting of over 4500 individuals. We analyzed the pathway composition of the latent space using a global feature importance score, which demonstrated that latent dimensions represent distinct cellular processes. To demonstrate model generalizability, we generated latent representations of unseen metabolomics datasets on type 2 diabetes, acute myeloid leukemia, and schizophrenia and found significant correlations with clinical patient groups. Notably, the VAE representations showed stronger effects than latent dimensions derived by linear and non-linear principal component analysis. Taken together, we demonstrate that the VAE is a powerful method that learns biologically meaningful, nonlinear, and transferrable latent representations of metabolomics data.

The impact of rare germline variants on human somatic mutation processes

Somatic mutations are an inevitable component of ageing and the most important cause of cancer. The rates and types of somatic mutation vary across individuals, but relatively few inherited influences on mutation processes are known. We perform a gene-based rare variant association study with diverse mutational processes, using human cancer genomes from over 11,000 individuals of European ancestry. By combining burden and variance tests, we identify 207 associations involving 15 somatic mutational phenotypes and 42 genes that replicated in an independent data set at a false discovery rate of 1%. We associate rare inherited deleterious variants in genes such as MSH3, EXO1, SETD2, and MTOR with two phenotypically different forms of DNA mismatch repair deficiency, and variants in genes such as EXO1, PAXIP1, RIF1, and WRN with deficiency in homologous recombination repair. In addition, we identify associations with other mutational processes, such as APEX1 with APOBEC-signature mutagenesis. Many of the genes interact with each other and with known mutator genes within cellular sub-networks. Considered collectively, damaging variants in the identified genes are prevalent in the population. We suggest that rare germline variation in diverse genes commonly impacts mutational processes in somatic cells.

Widespread redundancy in -omics profiles of cancer mutation states

BACKGROUND

In studies of cellular function in cancer, researchers are increasingly able to choose from many -omics assays as functional readouts. Choosing the correct readout for a given study can be difficult, and which layer of cellular function is most suitable to capture the relevant signal remains unclear.

RESULTS

We consider prediction of cancer mutation status (presence or absence) from functional -omics data as a representative problem that presents an opportunity to quantify and compare the ability of different -omics readouts to capture signals of dysregulation in cancer. From the TCGA Pan-Cancer Atlas that contains genetic alteration data, we focus on RNA sequencing, DNA methylation arrays, reverse phase protein arrays (RPPA), microRNA, and somatic mutational signatures as -omics readouts. Across a collection of genes recurrently mutated in cancer, RNA sequencing tends to be the most effective predictor of mutation state. We find that one or more other data types for many of the genes are approximately equally effective predictors. Performance is more variable between mutations than that between data types for the same mutation, and there is little difference between the top data types. We also find that combining data types into a single multi-omics model provides little or no improvement in predictive ability over the best individual data type.

CONCLUSIONS

Based on our results, for the design of studies focused on the functional outcomes of cancer mutations, there are often multiple -omics types that can serve as effective readouts, although gene expression seems to be a reasonable default option.

A multi-encoder variational autoencoder controls multiple transformational features in single-cell image analysis

Image-based cell phenotyping relies on quantitative measurements as encoded representations of cells; however, defining suitable representations that capture complex imaging features is challenged by the lack of robust methods to segment cells, identify subcellular compartments, and extract relevant features. Variational autoencoder (VAE) approaches produce encouraging results by mapping an image to a representative descriptor, and outperform classical hand-crafted features for morphology, intensity, and texture at differentiating data. Although VAEs show promising results for capturing morphological and organizational features in tissue, single cell image analyses based on VAEs often fail to identify biologically informative features due to uninformative technical variation. Here we propose a multi-encoder VAE (ME-VAE) in single cell image analysis using transformed images as a self-supervised signal to extract transform-invariant biologically meaningful features, including emergent features not obvious from prior knowledge. We show that the proposed architecture improves analysis by making distinct cell populations more separable compared to traditional and recent extensions of VAE architectures and intensity measurements by enhancing phenotypic differences between cells and by improving correlations to other analytic modalities. Better feature extraction and image analysis methods enabled by the ME-VAE will advance our understanding of complex cell biology and enable discoveries previously hidden behind image complexity ultimately improving medical outcomes and drug discovery.

Predicting drug polypharmacology from cell morphology readouts using variational autoencoder latent space arithmetic

A variational autoencoder (VAE) is a machine learning algorithm, useful for generating a compressed and interpretable latent space. These representations have been generated from various biomedical data types and can be used to produce realistic-looking simulated data. However, standard vanilla VAEs suffer from entangled and uninformative latent spaces, which can be mitigated using other types of VAEs such as β-VAE and MMD-VAE. In this project, we evaluated the ability of VAEs to learn cell morphology characteristics derived from cell images. We trained and evaluated these three VAE variants-Vanilla VAE, β-VAE, and MMD-VAE-on cell morphology readouts and explored the generative capacity of each model to predict compound polypharmacology (the interactions of a drug with more than one target) using an approach called latent space arithmetic (LSA). To test the generalizability of the strategy, we also trained these VAEs using gene expression data of the same compound perturbations and found that gene expression provides complementary information. We found that the β-VAE and MMD-VAE disentangle morphology signals and reveal a more interpretable latent space. We reliably simulated morphology and gene expression readouts from certain compounds thereby predicting cell states perturbed with compounds of known polypharmacology. Inferring cell state for specific drug mechanisms could aid researchers in developing and identifying targeted therapeutics and categorizing off-target effects in the future.

Optimal dimensionality selection for independent component analysis of transcriptomic data

Background

Independent component analysis is an unsupervised machine learning algorithm that separates a set of mixed signals into a set of statistically independent source signals. Applied to high-quality gene expression datasets, independent component analysis effectively reveals both the source signals of the transcriptome as co-regulated gene sets, and the activity levels of the underlying regulators across diverse experimental conditions. Two major variables that affect the final gene sets are the diversity of the expression profiles contained in the underlying data, and the user-defined number of independent components, or dimensionality, to compute. Availability of high-quality transcriptomic datasets has grown exponentially as high-throughput technologies have advanced; however, optimal dimensionality selection remains an open question.

Methods

We computed independent components across a range of dimensionalities for four gene expression datasets with varying dimensions (both in terms of number of genes and number of samples). We computed the correlation between independent components across different dimensionalities to understand how the overall structure evolves as the number of user-defined components increases. We then measured how well the resulting gene clusters reflected known regulatory mechanisms, and developed a set of metrics to assess the accuracy of the decomposition at a given dimension.

Results

We found that over-decomposition results in many independent components dominated by a single gene, whereas under-decomposition results in independent components that poorly capture the known regulatory structure. From these results, we developed a new method, called OptICA, for finding the optimal dimensionality that controls for both over- and under-decomposition. Specifically, OptICA selects the highest dimension that produces a low number of components that are dominated by a single gene. We show that OptICA outperforms two previously proposed methods for selecting the number of independent components across four transcriptomic databases of varying sizes.

Conclusions

OptICA avoids both over-decomposition and under-decomposition of transcriptomic datasets resulting in the best representation of the organism’s underlying transcriptional regulatory network.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12859-021-04497-7.

Charting Shifts in Saccharomyces cerevisiae Gene Expression across Asynchronous Time Trajectories with Diffusion Maps

During fermentation, Saccharomyces cerevisiae metabolizes sugars and other nutrients to obtain energy for growth and survival, while also modulating these activities in response to cell-environment interactions. Here, differences in S. cerevisiae gene expression were explored over a time course of fermentation and used to differentiate fermentations, using Pinot noir grapes from 15 unique sites. Data analysis was complicated by the fact that the fermentations proceeded at different rates, making a direct comparison of time series gene expression data difficult with conventional differential expression tools. This led to the development of a novel approach combining diffusion mapping with continuous differential expression analysis (termed DMap-DE). Using this method, site-specific deviations in gene expression were identified, including changes in gene expression correlated with the non-Saccharomyces yeast Hanseniaspora uvarum, as well as initial nitrogen concentrations in grape musts. These results highlight novel relationships between site-specific variables and Saccharomyces cerevisiae gene expression that are linked to repeated fermentation outcomes. It was also demonstrated that DMap-DE can extract biologically relevant gene expression patterns from other contexts (e.g., hypoxic response of Saccharomyces cerevisiae) and offers advantages over other data dimensionality reduction approaches, indicating that DMap-DE offers a robust method for investigating asynchronous time series gene expression data.

From bench to bedside: Single-cell analysis for cancer immunotherapy

Single-cell technologies are emerging as powerful tools for cancer research. These technologies characterize the molecular state of each cell within a tumor, enabling new exploration of tumor heterogeneity, microenvironment cell-type composition, and cell state transitions that affect therapeutic response, particularly in the context of immunotherapy. Analyzing clinical samples has great promise for precision medicine but is technically challenging. Successfully identifying predictors of response requires well-coordinated, multi-disciplinary teams to ensure adequate sample processing for high-quality data generation and computational analysis for data interpretation. Here, we review current approaches to sample processing and computational analysis regarding their application to translational cancer immunotherapy research.

Integrative Analysis of Next-Generation Sequencing for Next-Generation Cancer Research toward Artificial Intelligence

The rapid improvement of next-generation sequencing (NGS) technologies and their application in large-scale cohorts in cancer research led to common challenges of big data. It opened a new research area incorporating systems biology and machine learning. As large-scale NGS data accumulated, sophisticated data analysis methods became indispensable. In addition, NGS data have been integrated with systems biology to build better predictive models to determine the characteristics of tumors and tumor subtypes. Therefore, various machine learning algorithms were introduced to identify underlying biological mechanisms. In this work, we review novel technologies developed for NGS data analysis, and we describe how these computational methodologies integrate systems biology and omics data. Subsequently, we discuss how deep neural networks outperform other approaches, the potential of graph neural networks (GNN) in systems biology, and the limitations in NGS biomedical research. To reflect on the various challenges and corresponding computational solutions, we will discuss the following three topics: (i) molecular characteristics, (ii) tumor heterogeneity, and (iii) drug discovery. We conclude that machine learning and network-based approaches can add valuable insights and build highly accurate models. However, a well-informed choice of learning algorithm and biological network information is crucial for the success of each specific research question.

ACSNI: An unsupervised machine-learning tool for prediction of tissue-specific pathway components using gene expression profiles

Determining the tissue- and disease-specific circuit of biological pathways remains a fundamental goal of molecular biology. Many components of these biological pathways still remain unknown, hindering the full and accurate characterization of biological processes of interest. Here we describe ACSNI, an algorithm that combines prior knowledge of biological processes with a deep neural network to effectively decompose gene expression profiles (GEPs) into multi-variable pathway activities and identify unknown pathway components. Experiments on public GEP data show that ACSNI predicts cogent components of mTOR, ATF2, and HOTAIRM1 signaling that recapitulate regulatory information from genetic perturbation and transcription factor binding datasets. Our framework provides a fast and easy-to-use method to identify components of signaling pathways as a tool for molecular mechanism discovery and to prioritize genes for designing future targeted experiments (https://github.com/caanene1/ACSNI).

Forecasting cellular states: from descriptive to predictive biology via single-cell multiomics

As the single cell field races to characterize each cell type, state, and behavior, the complexity of the computational analysis approaches the complexity of the biological systems. Single cell and imaging technologies now enable unprecedented measurements of state transitions in biological systems, providing high-throughput data that capture tens-of-thousands of measurements on hundreds-of-thousands of samples. Thus, the definition of cell type and state is evolving to encompass the broad range of biological questions now attainable. To answer these questions requires the development of computational tools for integrated multi-omics analysis. Merged with mathematical models, these algorithms will be able to forecast future states of biological systems, going from statistical inferences of phenotypes to time course predictions of the biological systems with dynamic maps analogous to weather systems. Thus, systems biology for forecasting biological system dynamics from multi-omic data represents the future of cell biology empowering a new generation of technology-driven predictive medicine.

GeneWalk identifies relevant gene functions for a biological context using network representation learning

A bottleneck in high-throughput functional genomics experiments is identifying the most important genes and their relevant functions from a list of gene hits. Gene Ontology (GO) enrichment methods provide insight at the gene set level. Here, we introduce GeneWalk ( github.com/churchmanlab/genewalk ) that identifies individual genes and their relevant functions critical for the experimental setting under examination. After the automatic assembly of an experiment-specific gene regulatory network, GeneWalk uses representation learning to quantify the similarity between vector representations of each gene and its GO annotations, yielding annotation significance scores that reflect the experimental context. By performing gene- and condition-specific functional analysis, GeneWalk converts a list of genes into data-driven hypotheses.

Adversarial deconfounding autoencoder for learning robust gene expression embeddings

MOTIVATION

Increasing number of gene expression profiles has enabled the use of complex models, such as deep unsupervised neural networks, to extract a latent space from these profiles. However, expression profiles, especially when collected in large numbers, inherently contain variations introduced by technical artifacts (e.g. batch effects) and uninteresting biological variables (e.g. age) in addition to the true signals of interest. These sources of variations, called confounders, produce embeddings that fail to transfer to different domains, i.e. an embedding learned from one dataset with a specific confounder distribution does not generalize to different distributions. To remedy this problem, we attempt to disentangle confounders from true signals to generate biologically informative embeddings.

RESULTS

In this article, we introduce the Adversarial Deconfounding AutoEncoder (AD-AE) approach to deconfounding gene expression latent spaces. The AD-AE model consists of two neural networks: (i) an autoencoder to generate an embedding that can reconstruct original measurements, and (ii) an adversary trained to predict the confounder from that embedding. We jointly train the networks to generate embeddings that can encode as much information as possible without encoding any confounding signal. By applying AD-AE to two distinct gene expression datasets, we show that our model can (i) generate embeddings that do not encode confounder information, (ii) conserve the biological signals present in the original space and (iii) generalize successfully across different confounder domains. We demonstrate that AD-AE outperforms standard autoencoder and other deconfounding approaches.

AVAILABILITY AND IMPLEMENTATION

Our code and data are available at https://gitlab.cs.washington.edu/abdincer/ad-ae.

CONTACT

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

State of the Field in Multi-Omics Research: From Computational Needs to Data Mining and Sharing

Multi-omics, variously called integrated omics, pan-omics, and trans-omics, aims to combine two or more omics data sets to aid in data analysis, visualization and interpretation to determine the mechanism of a biological process. Multi-omics efforts have taken center stage in biomedical research leading to the development of new insights into biological events and processes. However, the mushrooming of a myriad of tools, datasets, and approaches tends to inundate the literature and overwhelm researchers new to the field. The aims of this review are to provide an overview of the current state of the field, inform on available reliable resources, discuss the application of statistics and machine/deep learning in multi-omics analyses, discuss findable, accessible, interoperable, reusable (FAIR) research, and point to best practices in benchmarking. Thus, we provide guidance to interested users of the domain by addressing challenges of the underlying biology, giving an overview of the available toolset, addressing common pitfalls, and acknowledging current methods' limitations. We conclude with practical advice and recommendations on software engineering and reproducibility practices to share a comprehensive awareness with new researchers in multi-omics for end-to-end workflow.

Responsible, practical genomic data sharing that accelerates research

Data sharing anchors reproducible science, but expectations and best practices are often nebulous. Communities of funders, researchers and publishers continue to grapple with what should be required or encouraged. To illuminate the rationales for sharing data, the technical challenges and the social and cultural challenges, we consider the stakeholders in the scientific enterprise. In biomedical research, participants are key among those stakeholders. Ethical sharing requires considering both the value of research efforts and the privacy costs for participants. We discuss current best practices for various types of genomic data, as well as opportunities to promote ethical data sharing that accelerates science by aligning incentives.

Latent Factor Modeling of scRNA-Seq Data Uncovers Dysregulated Pathways in Autoimmune Disease Patients

Latent factor modeling applied to single-cell RNA sequencing (scRNA-seq) data is a useful approach to discover gene signatures. However, it is often unclear what methods are best suited for specific tasks and how latent factors should be interpreted. Here, we compare four state-of-the-art methods and propose an approach to assign derived latent factors to pathway activities and specific cell subsets. By applying this framework to scRNA-seq datasets from biopsies of patients with rheumatoid arthritis and systemic lupus erythematosus, we discover disease-relevant gene signatures in specific cellular subsets. In rheumatoid arthritis, we identify an inflammatory OSMR signaling signature active in a subset of synovial fibroblasts and an efferocytic signature in a subset of synovial monocytes. Overall, we provide insights into latent factors models for the analysis of scRNA-seq data, develop a framework to identify cell subtypes in a phenotype-driven way, and use it to identify novel pathways dysregulated in rheumatoid arthritis.

Co-expression analysis reveals interpretable gene modules controlled by trans-acting genetic variants

Understanding the causal processes that contribute to disease onset and progression is essential for developing novel therapies. Although trans-acting expression quantitative trait loci (trans-eQTLs) can directly reveal cellular processes modulated by disease variants, detecting trans-eQTLs remains challenging due to their small effect sizes. Here, we analysed gene expression and genotype data from six blood cell types from 226 to 710 individuals. We used co-expression modules inferred from gene expression data with five methods as traits in trans-eQTL analysis to limit multiple testing and improve interpretability. In addition to replicating three established associations, we discovered a novel trans-eQTL near SLC39A8 regulating a module of metallothionein genes in LPS-stimulated monocytes. Interestingly, this effect was mediated by a transient cis-eQTL present only in early LPS response and lost before the trans effect appeared. Our analyses highlight how co-expression combined with functional enrichment analysis improves the identification and prioritisation of trans-eQTLs when applied to emerging cell-type-specific datasets.