An intrinsically interpretable neural network architecture for sequence-to-function learning.

MOTIVATION: Sequence-based deep learning approaches have been shown to predict a multitude of functional genomic readouts, including regions of open chromatin and RNA expression of genes. However, a major limitation of current methods is that model interpretation relies on computationally demanding post hoc analyses, and even then, one can often not explain the internal mechanics of highly parameterized models. Here, we introduce a deep learning architecture called totally interpretable sequence-to-function model (tiSFM). tiSFM improves upon the performance of standard multilayer convolutional models while using fewer parameters. Additionally, while tiSFM is itself technically a multilayer neural network, internal model parameters are intrinsically interpretable in terms of relevant sequence motifs. RESULTS: We analyze published open chromatin measurements across hematopoietic lineage cell-types and demonstrate that tiSFM outperforms a state-of-the-art convolutional neural network model custom-tailored to this dataset. We also show that it correctly identifies context-specific activities of transcription factors with known roles in hematopoietic differentiation, including Pax5 and Ebf1 for B-cells, and Rorc for innate lymphoid cells. tiSFM's model parameters have biologically meaningful interpretations, and we show the utility of our approach on a complex task of predicting the change in epigenetic state as a function of developmental transition. AVAILABILITY AND IMPLEMENTATION: The source code, including scripts for the analysis of key findings, can be found at https://github.com/boooooogey/ATAConv, implemented in Python.

An intrinsically interpretable neural network architecture for sequence to function learning.

MOTIVATION: Sequence-based deep learning approaches have been shown to predict a multitude of functional genomic readouts, including regions of open chromatin and RNA expression of genes. However, a major limitation of current methods is that model interpretation relies on computationally demanding post hoc analyses, and even then, one can often not explain the internal mechanics of highly parameterized models. Here, we introduce a deep learning architecture called tiSFM (totally interpretable sequence to function model). tiSFM improves upon the performance of standard multi-layer convolutional models while using fewer parameters. Additionally, while tiSFM is itself technically a multi-layer neural network, internal model parameters are intrinsically interpretable in terms of relevant sequence motifs. RESULTS: We analyze published open chromatin measurements across hematopoietic lineage cell-types and demonstrate that tiSFM outperforms a state-of-the-art convolutional neural network model custom-tailored to this dataset. We also show that it correctly identifies context specific activities of transcription factors with known roles in hematopoietic differentiation, including Pax5 and Ebf1 for B-cells, and Rorc for innate lymphoid cells. tiSFM's model parameters have biologically meaningful interpretations, and we show the utility of our approach on a complex task of predicting the change in epigenetic state as a function of developmental transition. AVAILABILITY AND IMPLEMENTATION: The source code, including scripts for the analysis of key findings, can be found at https://github.com/boooooogey/ATAConv, implemented in Python.

Apoptosis of hematopoietic progenitor-derived adipose tissue-resident macrophages contributes to insulin resistance after myocardial infarction.

Patients with insulin resistance have high risk of cardiovascular disease such as myocardial infarction (MI). However, it is not known whether MI can initiate or aggravate insulin resistance. We observed that patients with ST-elevation MI and mice with MI had de novo hyperglycemia and features of insulin resistance, respectively. In mouse models of both myocardial and skeletal muscle injury, we observed that the number of visceral adipose tissue (VAT)-resident macrophages decreased because of apoptosis after these distant organ injuries. Patients displayed a similar decrease in VAT-resident macrophage numbers and developed systemic insulin resistance after ST-elevation MI. Loss of VAT-resident macrophages after MI injury led to systemic insulin resistance in non-diabetic mice. Danger signaling-associated protein high mobility group box 1 was released by the dead myocardium after MI in rodents and triggered macrophage apoptosis via Toll-like receptor 4. The VAT-resident macrophage population in the steady state in mice was transcriptomically distinct from macrophages in the brain, skin, kidney, bone marrow, lungs, and liver and was derived from hematopoietic progenitor cells just after birth. Mechanistically, VAT-resident macrophage apoptosis and de novo insulin resistance in mouse models of MI were linked to diminished concentrations of macrophage colony-stimulating factor and adiponectin. Collectively, these findings demonstrate a previously unappreciated role of adipose tissue-resident macrophages in sensing remote organ injury and promoting MI pathogenesis.

FSTL-1 Attenuation Causes Spontaneous Smoke-Resistant Pulmonary Emphysema.

Rationale: The role of FSTL-1 (follistatin-like 1) in lung homeostasis is unknown.Objectives: We aimed to define the impact of FSTL-1 attenuation on lung structure and function and to identify FSTL-1-regulated transcriptional pathways in the lung. Further, we aimed to analyze the association of FSTL-1 SNPs with lung disease.Methods: FSTL-1 hypomorphic (FSTL-1 Hypo) mice underwent lung morphometry, pulmonary function testing, and micro-computed tomography. Fstl1 expression was determined in wild-type lung cell populations from three independent research groups. RNA sequencing of wild-type and FSTL-1 Hypo mice identified FSTL-1-regulated gene expression, followed by validation and mechanistic in vitro examination. FSTL1 SNP analysis was performed in the COPDGene (Genetic Epidemiology of Chronic Obstructive Pulmonary Disease) cohort.Measurements and Main Results: FSTL-1 Hypo mice developed spontaneous emphysema, independent of smoke exposure. Fstl1 is highly expressed in the lung by mesenchymal and endothelial cells but not immune cells. RNA sequencing of whole lung identified 33 FSTL-1-regulated genes, including Nr4a1, an orphan nuclear hormone receptor that negatively regulates NF-κB (nuclear factor-κB) signaling. In vitro, recombinant FSTL-1 treatment of macrophages attenuated NF-κB p65 phosphorylation in an Nr4a1-dependent manner. Within the COPDGene cohort, several SNPs in the FSTL1 region corresponded to chronic obstructive pulmonary disease and lung function.Conclusions: This work identifies a novel role for FSTL-1 protecting against emphysema development independent of smoke exposure. This FSTL-1-deficient emphysema implicates regulation of immune tolerance in lung macrophages through Nr4a1. Further study of the mechanisms involving FSTL-1 in lung homeostasis, immune regulation, and NF-κB signaling may provide additional insight into the pathophysiology of emphysema and inflammatory lung diseases.

Control of cytokinesis by ß-adrenergic receptors indicates an approach for regulating cardiomyocyte endowment.

One million patients with congenital heart disease (CHD) live in the United States. They have a lifelong risk of developing heart failure. Current concepts do not sufficiently address mechanisms of heart failure development specifically for these patients. Here, analysis of heart tissue from an infant with tetralogy of Fallot with pulmonary stenosis (ToF/PS) labeled with isotope-tagged thymidine demonstrated that cardiomyocyte cytokinesis failure is increased in this common form of CHD. We used single-cell transcriptional profiling to discover that the underlying mechanism of cytokinesis failure is repression of the cytokinesis gene ECT2, downstream of ß-adrenergic receptors (ß-ARs). Inactivation of the ß-AR genes and administration of the ß-blocker propranolol increased cardiomyocyte division in neonatal mice, which increased the number of cardiomyocytes (endowment) and conferred benefit after myocardial infarction in adults. Propranolol enabled the division of ToF/PS cardiomyocytes in vitro. These results suggest that ß-blockers could be evaluated for increasing cardiomyocyte division in patients with ToF/PS and other types of CHD.

Large-scale inference of competing endogenous RNA networks with sparse partial correlation.

MOTIVATION: MicroRNAs (miRNAs) are important non-coding post-transcriptional regulators that are involved in many biological processes and human diseases. Individual miRNAs may regulate hundreds of genes, giving rise to a complex gene regulatory network in which transcripts carrying miRNA binding sites act as competing endogenous RNAs (ceRNAs). Several methods for the analysis of ceRNA interactions exist, but these do often not adjust for statistical confounders or address the problem that more than one miRNA interacts with a target transcript. RESULTS: We present SPONGE, a method for the fast construction of ceRNA networks. SPONGE uses 'multiple sensitivity correlation', a newly defined measure for which we can estimate a distribution under a null hypothesis. SPONGE can accurately quantify the contribution of multiple miRNAs to a ceRNA interaction with a probabilistic model that addresses previously neglected confounding factors and allows fast P-value calculation, thus outperforming existing approaches. We applied SPONGE to paired miRNA and gene expression data from The Cancer Genome Atlas for studying global effects of miRNA-mediated cross-talk. Our results highlight already established and novel protein-coding and non-coding ceRNAs which could serve as biomarkers in cancer. AVAILABILITY AND IMPLEMENTATION: SPONGE is available as an R/Bioconductor package (doi: 10.18129/B9.bioc.SPONGE). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Von Hippel-Lindau Acts as a Metabolic Switch Controlling Nephron Progenitor Differentiation.

BACKGROUND: Nephron progenitors, the cell population that give rise to the functional unit of the kidney, are metabolically active and self-renew under glycolytic conditions. A switch from glycolysis to mitochondrial respiration drives these cells toward differentiation, but the mechanisms that control this switch are poorly defined. Studies have demonstrated that kidney formation is highly dependent on oxygen concentration, which is largely regulated by von Hippel-Lindau (VHL; a protein component of a ubiquitin ligase complex) and hypoxia-inducible factors (a family of transcription factors activated by hypoxia). METHODS: To explore VHL as a regulator defining nephron progenitor self-renewal versus differentiation, we bred Six2-TGCtg mice with VHLlox/lox mice to generate mice with a conditional deletion of VHL from Six2+ nephron progenitors. We used histologic, immunofluorescence, RNA sequencing, and metabolic assays to characterize kidneys from these mice and controls during development and up to postnatal day 21. RESULTS: By embryonic day 15.5, kidneys of nephron progenitor cell-specific VHL knockout mice begin to exhibit reduced maturation of nephron progenitors. Compared with controls, VHL knockout kidneys are smaller and developmentally delayed by postnatal day 1, and have about half the number of glomeruli at postnatal day 21. VHL knockout nephron progenitors also exhibit persistent Six2 and Wt1 expression, as well as decreased mitochondrial respiration and prolonged reliance on glycolysis. CONCLUSIONS: Our findings identify a novel role for VHL in mediating nephron progenitor differentiation through metabolic regulation, and suggest that VHL is required for normal kidney development.

Loss of miR-17~92 results in dysregulation of Cftr in nephron progenitors.

We have previously demonstrated that loss of miR-17~92 in nephron progenitors in a mouse model results in renal hypodysplasia and chronic kidney disease. Clinically, decreased congenital nephron endowment because of renal hypodysplasia is associated with an increased risk of hypertension and chronic kidney disease, and this is at least partly dependent on the self-renewal of nephron progenitors. Here, we present evidence for a novel molecular mechanism regulating the self-renewal of nephron progenitors and congenital nephron endowment by the highly conserved miR-17~92 cluster. Whole transcriptome sequencing revealed that nephron progenitors lacking this cluster demonstrated increased Cftr expression. We showed that one member of the cluster, miR-19b, is sufficient to repress Cftr expression in vitro and that perturbation of Cftr activity in nephron progenitors results in impaired proliferation. Together, these data suggest that miR-19b regulates Cftr expression in nephron progenitors, with this interaction playing a role in appropriate nephron progenitor self-renewal during kidney development to generate normal nephron endowment.

Regulatory networks define phenotypic classes of human stem cell lines.

Stem cells are defined as self-renewing cell populations that can differentiate into multiple distinct cell types. However, hundreds of different human cell lines from embryonic, fetal and adult sources have been called stem cells, even though they range from pluripotent cells-typified by embryonic stem cells, which are capable of virtually unlimited proliferation and differentiation-to adult stem cell lines, which can generate a far more limited repertoire of differentiated cell types. The rapid increase in reports of new sources of stem cells and their anticipated value to regenerative medicine has highlighted the need for a general, reproducible method for classification of these cells. We report here the creation and analysis of a database of global gene expression profiles (which we call the 'stem cell matrix') that enables the classification of cultured human stem cells in the context of a wide variety of pluripotent, multipotent and differentiated cell types. Using an unsupervised clustering method to categorize a collection of approximately 150 cell samples, we discovered that pluripotent stem cell lines group together, whereas other cell types, including brain-derived neural stem cell lines, are very diverse. Using further bioinformatic analysis we uncovered a protein-protein network (PluriNet) that is shared by the pluripotent cells (embryonic stem cells, embryonal carcinomas and induced pluripotent cells). Analysis of published data showed that the PluriNet seems to be a common characteristic of pluripotent cells, including mouse embryonic stem and induced pluripotent cells and human oocytes. Our results offer a new strategy for classifying stem cells and support the idea that pluripotency and self-renewal are under tight control by specific molecular networks.

Detecting hierarchical structure in molecular characteristics of disease using transitive approximations of directed graphs.

MOTIVATION: Molecular diagnostics aims at classifying diseases into clinically relevant sub-entities based on molecular characteristics. Typically, the entities are split into subgroups, which might contain several variants yielding a hierarchical model of the disease. Recent years have introduced a plethora of new molecular screening technologies to molecular diagnostics. As a result molecular profiles of patients became complex and the classification task more difficult. RESULTS: We present a novel tool for detecting hierarchical structure in binary datasets. We aim for identifying molecular characteristics, which are stochastically implying other characteristics. The final hierarchical structure is encoded in a directed transitive graph where nodes represent molecular characteristics and a directed edge from a node A to a node B denotes that almost all cases with characteristic B also display characteristic A. Naturally, these graphs need to be transitive. In the core of our modeling approach lies the problem of calculating good transitive approximations of given directed but not necessarily transitive graphs. By good transitive approximation we understand transitive graphs, which differ from the reference graph in only a small number of edges. It is known that the problem of finding optimal transitive approximation is NP-complete. Here we develop an efficient heuristic for generating good transitive approximations. We evaluate the computational efficiency of the algorithm in simulations, and demonstrate its use in the context of a large genome-wide study on mature aggressive lymphomas. AVAILABILITY: The software used in our analysis is freely available from http://compdiag.uni-regensburg.de/software/transApproxs.shtml.

Microarray based diagnosis profits from better documentation of gene expression signatures.

Microarray gene expression signatures hold great promise to improve diagnosis and prognosis of disease. However, current documentation standards of such signatures do not allow for an unambiguous application to study-external patients. This hinders independent evaluation, effectively delaying the use of signatures in clinical practice. Data from eight publicly available clinical microarray studies were analyzed and the consistency of study-internal with study-external diagnoses was evaluated. Study-external classifications were based on documented information only. Documenting a signature is conceptually different from reporting a list of genes. We show that even the exact quantitative specification of a classification rule alone does not define a signature unambiguously. We found that discrepancy between study-internal and study-external diagnoses can be as frequent as 30% (worst case) and 18% (median). By using the proposed documentation by value strategy, which documents quantitative preprocessing information, the median discrepancy was reduced to 1%. The process of evaluating microarray gene expression diagnostic signatures and bringing them to clinical practice can be substantially improved and made more reliable by better documentation of the signatures.

Expression of late cell cycle genes and an increased proliferative capacity characterize very early relapse of childhood acute lymphoblastic leukemia.

PURPOSE: In childhood acute lymphoblastic leukemia (ALL), approximately 25% of patients suffer from relapse. In recurrent disease, despite intensified therapy, overall cure rates of 40% remain unsatisfactory and survival rates are particularly poor in certain subgroups. The probability of long-term survival after relapse is predicted from well-established prognostic factors (i.e., time and site of relapse, immunophenotype, and minimal residual disease). However, the underlying biological determinants of these prognostic factors remain poorly understood. EXPERIMENTAL DESIGN: Aiming at identifying molecular pathways associated with these clinically well-defined prognostic factors, we did gene expression profiling on 60 prospectively collected samples of first relapse patients enrolled on the relapse trial ALL-REZ BFM 2002 of the Berlin-Frankfurt-Münster study group. RESULTS: We show here that patients with very early relapse of ALL are characterized by a distinctive gene expression pattern. We identified a set of 83 genes differentially expressed in very early relapsed ALL compared with late relapsed disease. The vast majority of genes were up-regulated and many were late cell cycle genes with a function in mitosis. In addition, samples from patients with very early relapse showed a significant increase in the percentage of S and G(2)-M phase cells and this correlated well with the expression level of cell cycle genes. CONCLUSIONS: Very early relapse of ALL is characterized by an increased proliferative capacity of leukemic blasts and up-regulated mitotic genes. The latter suggests that novel drugs, targeting late cell cycle proteins, might be beneficial for these patients that typically face a dismal prognosis.

Finding disease specific alterations in the co-expression of genes.

MOTIVATION: Standard analysis routines for microarray data aim at differentially expressed genes. In this paper, we address the complementary problem of detecting sets of differentially co-expressed genes in two phenotypically distinct sets of expression profiles. RESULTS: We introduce a score for differential co-expression and suggest a computationally efficient algorithm for finding high scoring sets of genes. The use of our novel method is demonstrated in the context of simulations and on real expression data from a clinical study.