A deep learning framework for denoising and ordering scRNA-seq data using adversarial autoencoder with dynamic batching.

Single-cell RNA sequencing (scRNA-seq) provides high resolution of cell-to-cell variation in gene expression and offers insights into cell heterogeneity, differentiating dynamics, and disease mechanisms. However, technical challenges such as low capture rates and dropout events can introduce noise in data analysis. Here, we present a deep learning framework, called the dynamic batching adversarial autoencoder (DB-AAE), for denoising scRNA-seq datasets. First, we describe steps to set up the computing environment, training, and tuning. Then, we depict the visualization of the denoising results. For complete details on the use and execution of this protocol, please refer to Ko et al.1.

Psat1-generated α-ketoglutarate and glutamine promote muscle stem cell activation and regeneration.

By satisfying bioenergetic demands, generating biomass, and providing metabolites serving as cofactors for chromatin modifiers, metabolism regulates adult stem cell biology. Here, we report that a branch of glycolysis, the serine biosynthesis pathway (SBP), is activated in regenerating muscle stem cells (MuSCs). Gene inactivation and metabolomics revealed that Psat1, one of the three SBP enzymes, controls MuSC activation and expansion of myogenic progenitors through production of the metabolite α-ketoglutarate (α-KG) and α-KG-generated glutamine. Psat1 ablation resulted in defective expansion of MuSCs and impaired regeneration. Psat1, α-KG, and glutamine were reduced in MuSCs of old mice. α-KG or glutamine re-established appropriate muscle regeneration of adult conditional Psat1 -/- mice and of old mice. These findings contribute insights into the metabolic role of Psat1 during muscle regeneration and suggest α-KG and glutamine as potential therapeutic interventions to ameliorate muscle regeneration during aging.

A deep learning adversarial autoencoder with dynamic batching displays high performance in denoising and ordering scRNA-seq data.

By providing high-resolution of cell-to-cell variation in gene expression, single-cell RNA sequencing (scRNA-seq) offers insights into cell heterogeneity, differentiating dynamics, and disease mechanisms. However, challenges such as low capture rates and dropout events can introduce noise in data analysis. Here, we propose a deep neural generative framework, the dynamic batching adversarial autoencoder (DB-AAE), which excels at denoising scRNA-seq datasets. DB-AAE directly captures optimal features from input data and enhances feature preservation, including cell type-specific gene expression patterns. Comprehensive evaluation on simulated and real datasets demonstrates that DB-AAE outperforms other methods in denoising accuracy and biological signal preservation. It also improves the accuracy of other algorithms in establishing pseudo-time inference. This study highlights DB-AAE's effectiveness and potential as a valuable tool for enhancing the quality and reliability of downstream analyses in scRNA-seq research.

Integrating single-cell transcriptomes, chromatin accessibility, and multiomics analysis of mesoderm-induced embryonic stem cells.

Here, we present workflows for integrating independent transcriptomic and chromatin accessibility datasets and analyzing multiomics. First, we describe steps for integrating independent transcriptomic and chromatin accessibility measurements. Next, we detail multimodal analysis of transcriptomes and chromatin accessibility performed in the same sample. We demonstrate their use by analyzing datasets obtained from mouse embryonic stem cells induced to differentiate toward mesoderm-like, myogenic, or neurogenic phenotypes. For complete details on the use and execution of this protocol, please refer to Khateb et al.1.

Polycomb Ezh1 maintains murine muscle stem cell quiescence through non-canonical regulation of Notch signaling.

Organismal homeostasis and regeneration are predicated on committed stem cells that can reside for long periods in a mitotically dormant but reversible cell-cycle arrest state defined as quiescence. Premature escape from quiescence is detrimental, as it results in stem cell depletion, with consequent defective tissue homeostasis and regeneration. Here, we report that Polycomb Ezh1 confers quiescence to murine muscle stem cells (MuSCs) through a non-canonical function. In the absence of Ezh1, MuSCs spontaneously exit quiescence. Following repeated injuries, the MuSC pool is progressively depleted, resulting in failure to sustain proper muscle regeneration. Rather than regulating repressive histone H3K27 methylation, Ezh1 maintains gene expression of the Notch signaling pathway in MuSCs. Selective genetic reconstitution of the Notch signaling corrects stem cell number and re-establishes quiescence of Ezh1-/- MuSCs.

Transcriptomics, regulatory syntax, and enhancer identification in mesoderm-induced ESCs at single-cell resolution.

Embryonic stem cells (ESCs) can adopt lineage-specific gene-expression programs by stepwise exposure to defined factors, resulting in the generation of functional cell types. Bulk and single-cell-based assays were employed to catalog gene expression, histone modifications, chromatin conformation, and accessibility transitions in ESC populations and individual cells acquiring a presomitic mesoderm fate and undergoing further specification toward myogenic and neurogenic lineages. These assays identified cis-regulatory regions and transcription factors presiding over gene-expression programs occurring at defined ESC transitions and revealed the presence of heterogeneous cell populations within discrete ESC developmental stages. The datasets were employed to identify previously unappreciated genomic elements directing the initial activation of Pax7 and myogenic and neurogenic gene-expression programs. This study provides a resource for the discovery of genomic and transcriptional features of pluripotent, mesoderm-induced ESCs and ESC-derived cell lineages.

Single cell analysis of adult mouse skeletal muscle stem cells in homeostatic and regenerative conditions.

Dedicated stem cells ensure postnatal growth, repair and homeostasis of skeletal muscle. Following injury, muscle stem cells (MuSCs) exit from quiescence and divide to reconstitute the stem cell pool and give rise to muscle progenitors. The transcriptomes of pooled MuSCs have provided a rich source of information for describing the genetic programs of distinct static cell states; however, bulk microarray and RNA sequencing provide only averaged gene expression profiles, blurring the heterogeneity and developmental dynamics of asynchronous MuSC populations. Instead, the granularity required to identify distinct cell types, states, and their dynamics can be afforded by single cell analysis. We were able to compare the transcriptomes of thousands of MuSCs and primary myoblasts isolated from homeostatic or regenerating muscles by single cell RNA sequencing. Using computational approaches, we could reconstruct dynamic trajectories and place, in a pseudotemporal manner, the transcriptomes of individual MuSC within these trajectories. This approach allowed for the identification of distinct clusters of MuSCs and primary myoblasts with partially overlapping but distinct transcriptional signatures, as well as the description of metabolic pathways associated with defined MuSC states.

A Muscle-Specific Enhancer RNA Mediates Cohesin Recruitment and Regulates Transcription In trans.

The enhancer regions of the myogenic master regulator MyoD give rise to at least two enhancer RNAs. Core enhancer eRNA (CEeRNA) regulates transcription of the adjacent MyoD gene, whereas DRReRNA affects expression of Myogenin in trans. We found that DRReRNA is recruited at the Myogenin locus, where it colocalizes with Myogenin nascent transcripts. DRReRNA associates with the cohesin complex, and this association correlates with its transactivating properties. Despite being expressed in undifferentiated cells, cohesin is not loaded on Myogenin until the cells start expressing DRReRNA, which is then required for cohesin chromatin recruitment and maintenance. Functionally, depletion of either cohesin or DRReRNA reduces chromatin accessibility, prevents Myogenin activation, and hinders muscle cell differentiation. Thus, DRReRNA ensures spatially appropriate cohesin loading in trans to regulate gene expression.

The Elongation Factor Spt6 Maintains ESC Pluripotency by Controlling Super-Enhancers and Counteracting Polycomb Proteins.

Spt6 coordinates nucleosome dis- and re-assembly, transcriptional elongation, and mRNA processing. Here, we report that depleting Spt6 in embryonic stem cells (ESCs) reduced expression of pluripotency factors, increased expression of cell-lineage-affiliated developmental regulators, and induced cell morphological and biochemical changes indicative of ESC differentiation. Selective downregulation of pluripotency factors upon Spt6 depletion may be mechanistically explained by its enrichment at ESC super-enhancers, where Spt6 controls histone H3K27 acetylation and methylation and super-enhancer RNA transcription. In ESCs, Spt6 interacted with the PRC2 core subunit Suz12 and prevented H3K27me3 accumulation at ESC super-enhancers and associated promoters. Biochemical as well as functional experiments revealed that Spt6 could compete for binding of the PRC2 methyltransferase Ezh2 to Suz12 and reduce PRC2 chromatin engagement. Thus, in addition to serving as a histone chaperone and transcription elongation factor, Spt6 counteracts repression by opposing H3K27me3 deposition at critical genomic regulatory regions.

Roles of H3K27me2 and H3K27me3 Examined during Fate Specification of Embryonic Stem Cells.

The polycomb repressive complex 2 (PRC2) methylates lysine 27 of histone H3 (H3K27) through its catalytic subunit Ezh2. PRC2-mediated di- and tri-methylation (H3K27me2/H3K27me3) have been interchangeably associated with gene repression. However, it remains unclear whether these two degrees of H3K27 methylation have different functions. In this study, we have generated isogenic mouse embryonic stem cells (ESCs) with a modified H3K27me2/H3K27me3 ratio. Our findings document dynamic developmental control in the genomic distribution of H3K27me2 and H3K27me3 at regulatory regions in ESCs. They also reveal that modifying the ratio of H3K27me2 and H3K27me3 is sufficient for the acquisition and repression of defined cell lineage transcriptional programs and phenotypes and influences induction of the ESC ground state.

Polycomb Ezh2 controls the fate of GABAergic neurons in the embryonic cerebellum.

Although the genetic interactions between signaling pathways and transcription factors have been largely decoded, much remains to be learned about the epigenetic regulation of cerebellar development. Here, we report that cerebellar deletion of Ezh2, the methyltransferase subunit of the PRC2 complex, results in reduced H3K27me3 and profound transcriptional dysregulation, including that of a set of transcription factors directly involved in cerebellar neuronal cell-type specification and differentiation. Such transcriptional changes lead to increased GABAergic interneurons and decreased Purkinje cells. Transcriptional changes also inhibit the proliferation of granule precursor cells derived from the rhombic lip. The loss of both cell types ultimately results in cerebellar hypoplasia. These findings indicate Ezh2/PRC2 plays crucial roles in regulating neurogenesis from both cerebellar germinal zones.

Predicting the severity of motor neuron disease progression using electronic health record data with a cloud computing Big Data approach.

Motor neuron diseases (MNDs) are a class of progressive neurological diseases that damage the motor neurons. An accurate diagnosis is important for the treatment of patients with MNDs because there is no standard cure for the MNDs. However, the rates of false positive and false negative diagnoses are still very high in this class of diseases. In the case of Amyotrophic Lateral Sclerosis (ALS), current estimates indicate 10% of diagnoses are false-positives, while 44% appear to be false negatives. In this study, we developed a new methodology to profile specific medical information from patient medical records for predicting the progression of motor neuron diseases. We implemented a system using Hbase and the Random forest classifier of Apache Mahout to profile medical records provided by the Pooled Resource Open-Access ALS Clinical Trials Database (PRO-ACT) site, and we achieved 66% accuracy in the prediction of ALS progress.

PHYRN: a robust method for phylogenetic analysis of highly divergent sequences.

Both multiple sequence alignment and phylogenetic analysis are problematic in the "twilight zone" of sequence similarity (≤ 25% amino acid identity). Herein we explore the accuracy of phylogenetic inference at extreme sequence divergence using a variety of simulated data sets. We evaluate four leading multiple sequence alignment (MSA) methods (MAFFT, T-COFFEE, CLUSTAL, and MUSCLE) and six commonly used programs of tree estimation (Distance-based: Neighbor-Joining; Character-based: PhyML, RAxML, GARLI, Maximum Parsimony, and Bayesian) against a novel MSA-independent method (PHYRN) described here. Strikingly, at "midnight zone" genetic distances (~7% pairwise identity and 4.0 gaps per position), PHYRN returns high-resolution phylogenies that outperform traditional approaches. We reason this is due to PHRYN's capability to amplify informative positions, even at the most extreme levels of sequence divergence. We also assess the applicability of the PHYRN algorithm for inferring deep evolutionary relationships in the divergent DANGER protein superfamily, for which PHYRN infers a more robust tree compared to MSA-based approaches. Taken together, these results demonstrate that PHYRN represents a powerful mechanism for mapping uncharted frontiers in highly divergent protein sequence data sets.

Predicting protein folds with fold-specific PSSM libraries.

Accurately assigning folds for divergent protein sequences is a major obstacle to structural studies. Herein, we outline an effective method for fold recognition using sets of PSSMs, each of which is constructed for different protein folds. Our analyses demonstrate that FSL (Fold-specific Position Specific Scoring Matrix Libraries) can predict/relate structures given only their amino acid sequences of highly divergent proteins. This ability to detect distant relationships is dependent on low-identity sequence alignments obtained from FSL. Results from our experiments demonstrate that FSL perform well in recognizing folds from the "twilight-zone" SABmark dataset. Further, this method is capable of accurate fold prediction in newly determined structures. We suggest that by building complete PSSM libraries for all unique folds within the Protein Database (PDB), FSL can be used to rapidly and reliably annotate a large subset of protein folds at proteomic level. The related programs and fold-specific PSSMs for our FSL are publicly available at: http://ccp.psu.edu/download/FSLv1.0/.

Phylogenetic Profiles Reveal Structural and Functional Determinants of Lipid-binding.

One of the major challenges in the genomic era is annotating structure/function to the vast quantities of sequence information now available. Indeed, most of the protein sequence database lacks comprehensive annotation, even when experimental evidence exists. Further, within structurally resolved and functionally annotated protein domains, additional functionalities contained in these domains are not apparent. To add further complication, small changes in the amino-acid sequence can lead to profound changes in both structure and function, underscoring the need for rapid and reliable methods to analyze these types of data. Phylogenetic profiles provide a quantitative method that can relate the structural and functional properties of proteins, as well as their evolutionary relationships. Using all of the structurally resolved Src-Homology-2 (SH2) domains, we demonstrate that knowledge-bases can be used to create single-amino acid phylogenetic profiles which reliably annotate lipid-binding. Indeed, these measures isolate the known phosphotyrosine and hydrophobic pockets as integral to lipid-binding function. In addition, we determined that the SH2 domain of Tec family kinases bind to lipids with varying affinity and specificity. Simulating mutations in Bruton's tyrosine kinase (BTK) that cause X-Linked Agammaglobulinemia (XLA) predict that these mutations alter lipid-binding, which we confirm experimentally. In light of these results, we propose that XLA-causing mutations in the SH3-SH2 domain of BTK alter lipid-binding, which could play a causative role in the XLA-phenotype. Overall, our study suggests that the number of lipid-binding proteins is drastically underestimated and, with further development, phylogenetic profiles can provide a method for rapidly increasing the functional annotation of protein sequences.

Phylogenetic profiles reveal structural/functional determinants of TRPC3 signal-sensing antennae.

Biochemical assessment of channel structure/function is incredibly challenging. Developing computational tools that provide these data would enable translational research, accelerating mechanistic experimentation for the bench scientist studying ion channels. Starting with the premise that protein sequence encodes information about structure, function and evolution (SF&E), we developed a unified framework for inferring SF&E from sequence information using a knowledge-based approach. The Gestalt Domain Detection Algorithm-Basic Local Alignment Tool (GDDA-BLAST) provides phylogenetic profiles that can model, ab initio, SF&E relationships of biological sequences at the whole protein, single domain and single-amino acid level.1,2 In our recent paper,4 we have applied GDDA-BLAST analysis to study canonical TRP (TRPC) channels1 and empirically validated predicted lipid-binding and trafficking activities contained within the TRPC3 TRP_2 domain of unknown function. Overall, our in silico, in vitro, and in vivo experiments support a model in which TRPC3 has signal-sensing antennae which are adorned with lipid-binding, trafficking and calmodulin regulatory domains. In this Addendum, we correlate our functional domain analysis with the cryo-EM structure of TRPC3.3 In addition, we synthesize recent studies with our new findings to provide a refined model on the mechanism(s) of TRPC3 activation/deactivation.

Phylogenetic profiles reveal evolutionary relationships within the "twilight zone" of sequence similarity.

Inferring evolutionary relationships among highly divergent protein sequences is a daunting task. In particular, when pairwise sequence alignments between protein sequences fall <25% identity, the phylogenetic relationships among sequences cannot be estimated with statistical certainty. Here, we show that phylogenetic profiles generated with the Gestalt Domain Detection Algorithm-Basic Local Alignment Tool (GDDA-BLAST) are capable of deriving, ab initio, phylogenetic relationships for highly divergent proteins in a quantifiable and robust manner. Notably, the results from our computational case study of the highly divergent family of retroelements accord with previous estimates of their evolutionary relationships. Taken together, these data demonstrate that GDDA-BLAST provides an independent and powerful measure of evolutionary relationships that does not rely on potentially subjective sequence alignment. We demonstrate that evolutionary relationships can be measured with phylogenetic profiles, and therefore propose that these measurements can provide key insights into relationships among distantly related and/or rapidly evolving proteins.