Integrating a Multi-label Deep Learning Approach with Protein Information to Compare Bioactive Peptides in Brain and Plasma.

Peptide therapeutics is gaining momentum. Advances in the field of peptidomics have enabled researchers to harvest vital information from various organisms and tissue types concerning peptide existence, expression and function. The development of mass spectrometry techniques for high-throughput peptide quantitation has paved the way for the identification and discovery of numerous known and novel peptides. Though much has been achieved, scientists are still facing difficulties when it comes to reducing the search space of the large mass spectrometry-generated peptidomics datasets and focusing on the subset of functionally relevant peptides. Moreover, there is currently no straightforward way to analytically compare the distributions of bioactive peptides in distinct biological samples, which may reveal much useful information when seeking to characterize tissue- or fluid-specific peptidomes. In this chapter, we demonstrate how to identify, rank, and compare predicted bioactive peptides and bioactivity distributions from extensive peptidomics datasets. To aid this task, we utilize MultiPep, a multi-label deep learning approach designed for classifying peptide bioactivities, to identify bioactive peptides. The predicted bioactivities are synergistically combined with protein information from the UniProt database, which assist in navigating through the jungle of putative therapeutic peptides and relevant peptide leads.

Adipogenic and SWAT cells separate from a common progenitor in human brown and white adipose depots.

Adipocyte function is a major determinant of metabolic disease, warranting investigations of regulating mechanisms. We show at single-cell resolution that progenitor cells from four human brown and white adipose depots separate into two main cell fates, an adipogenic and a structural branch, developing from a common progenitor. The adipogenic gene signature contains mitochondrial activity genes, and associates with genome-wide association study traits for fat distribution. Based on an extracellular matrix and developmental gene signature, we name the structural branch of cells structural Wnt-regulated adipose tissue-resident (SWAT) cells. When stripped from adipogenic cells, SWAT cells display a multipotent phenotype by reverting towards progenitor state or differentiating into new adipogenic cells, dependent on media. Label transfer algorithms recapitulate the cell types in human adipose tissue datasets. In conclusion, we provide a differentiation map of human adipocytes and define the multipotent SWAT cell, providing a new perspective on adipose tissue regulation.

De-novo reconstruction and identification of transcriptional gene regulatory network modules differentiating single-cell clusters.

Single-cell RNA sequencing (scRNA-seq) technology provides an unprecedented opportunity to understand gene functions and interactions at single-cell resolution. While computational tools for scRNA-seq data analysis to decipher differential gene expression profiles and differential pathway expression exist, we still lack methods to learn differential regulatory disease mechanisms directly from the single-cell data. Here, we provide a new methodology, named DiNiro, to unravel such mechanisms de novo and report them as small, easily interpretable transcriptional regulatory network modules. We demonstrate that DiNiro is able to uncover novel, relevant, and deep mechanistic models that not just predict but explain differential cellular gene expression programs. DiNiro is available at https://exbio.wzw.tum.de/diniro/.

MultiPep: a hierarchical deep learning approach for multi-label classification of peptide bioactivities.

Peptide-based therapeutics are here to stay and will prosper in the future. A key step in identifying novel peptide-drugs is the determination of their bioactivities. Recent advances in peptidomics screening approaches hold promise as a strategy for identifying novel drug targets. However, these screenings typically generate an immense number of peptides and tools for ranking these peptides prior to planning functional studies are warranted. Whereas a couple of tools in the literature predict multiple classes, these are constructed using multiple binary classifiers. We here aimed to use an innovative deep learning approach to generate an improved peptide bioactivity classifier with capacity of distinguishing between multiple classes. We present MultiPep: a deep learning multi-label classifier that assigns peptides to zero or more of 20 bioactivity classes. We train and test MultiPep on data from several publically available databases. The same data are used for a hierarchical clustering, whose dendrogram shapes the architecture of MultiPep. We test a new loss function that combines a customized version of Matthews correlation coefficient with binary cross entropy (BCE), and show that this is better than using class-weighted BCE as loss function. Further, we show that MultiPep surpasses state-of-the-art peptide bioactivity classifiers and that it predicts known and novel bioactivities of FDA-approved therapeutic peptides. In conclusion, we present innovative machine learning techniques used to produce a peptide prediction tool to aid peptide-based therapy development and hypothesis generation.

Network Medicine-Based Unbiased Disease Modules for Drug and Diagnostic Target Identification in ROSopathies.

Most diseases are defined by a symptom, not a mechanism. Consequently, therapies remain symptomatic. In reverse, many potential disease mechanisms remain in arbitrary search for clinical relevance. Reactive oxygen species (ROS) are such an example. It is an attractive hypothesis that dysregulation of ROS can become a disease trigger. Indeed, elevated ROS levels of various biomarkers have been correlated with almost every disease, yet after decades of research without any therapeutic application. We here present a first systematic, non-hypothesis-based approach to transform this field as a proof of concept for biomedical research in general. We selected as seed proteins 9 families with 42 members of clinically researched ROS-generating enzymes, ROS-metabolizing enzymes or ROS targets. Applying an unbiased network medicine approach, their first neighbours were connected, and, based on a stringent subnet participation degree (SPD) of 0.4, hub nodes excluded. This resulted in 12 distinct human interactome-based ROS signalling modules, while 8 proteins remaining unconnected. This ROSome is in sharp contrast to commonly used highly curated and integrated KEGG, HMDB or WikiPathways. These latter serve more as mind maps of possible ROS signalling events but may lack important interactions and often do not take different cellular and subcellular localization into account. Moreover, novel non-ROS-related proteins were part of these forming functional hybrids, such as the NOX5/sGC, NOX1,2/NOS2, NRF2/ENC-1 and MPO/SP-A modules. Thus, ROS sources are not interchangeable but associated with distinct disease processes or not at all. Module members represent leads for precision diagnostics to stratify patients with specific ROSopathies for precision intervention. The upper panel shows the classical approach to generate hypotheses for a role of ROS in a given disease by focusing on ROS levels and to some degree the ROS type or metabolite. Low levels are considered physiological; higher amounts are thought to cause a redox imbalance, oxidative stress and eventually disease. The source of ROS is less relevant; there is also ROS-induced ROS formation, i.e. by secondary sources (see upwards arrow). The non-hypothesis-based network medicine approach uses genetically or otherwise validated risk genes to construct disease-relevant signalling modules, which will contain also ROS targets. Not all ROS sources will be relevant for a given disease; some may not be disease relevant at all. The three examples show (from left to right) the disease-relevant appearance of an unphysiological ROS modifier/toxifier protein, ROS target or ROS source.

Enabling single-cell trajectory network enrichment.

Single-cell sequencing (scRNA-seq) technologies allow the investigation of cellular differentiation processes with unprecedented resolution. Although powerful software packages for scRNA-seq data analysis exist, systems biology-based tools for trajectory analysis are rare and typically difficult to handle. This hampers biological exploration and prevents researchers from gaining deeper insights into the molecular control of developmental processes. Here, to address this, we have developed Scellnetor; a network-constraint time-series clustering algorithm. It allows extraction of temporal differential gene expression network patterns (modules) that explain the difference in regulation of two developmental trajectories. Using well-characterized experimental model systems, we demonstrate the capacity of Scellnetor as a hypothesis generator to identify putative mechanisms driving haematopoiesis or mechanistically interpretable subnetworks driving dysfunctional CD8 T-cell development in chronic infections. Altogether, Scellnetor allows for single-cell trajectory network enrichment, which effectively lifts scRNA-seq data analysis to a systems biology level.

The Selection of a Hepatocyte Cell Line Susceptible to Plasmodium falciparum Sporozoite Invasion That Is Associated With Expression of Glypican-3.

In vitro studies of liver stage (LS) development of the human malaria parasite Plasmodium falciparum are technically challenging; therefore, fundamental questions about hepatocyte receptors for invasion that can be targeted to prevent infection remain unanswered. To identify novel receptors and to further understand human hepatocyte susceptibility to P. falciparum sporozoite invasion, we created an optimized in vitro system by mimicking in vivo liver conditions and using the subcloned HC-04.J7 cell line that supports mean infection rates of 3-5% and early development of P. falciparum exoerythrocytic forms-a 3- to 5-fold improvement on current in vitro hepatocarcinoma models for P. falciparum invasion. We juxtaposed this invasion-susceptible cell line with an invasion-resistant cell line (HepG2) and performed comparative proteomics and RNA-seq analyses to identify host cell surface molecules and pathways important for sporozoite invasion of host cells. We identified and investigated a hepatocyte cell surface heparan sulfate proteoglycan, glypican-3, as a putative mediator of sporozoite invasion. We also noted the involvement of pathways that implicate the importance of the metabolic state of the hepatocyte in supporting LS development. Our study highlights important features of hepatocyte biology, and specifically the potential role of glypican-3, in mediating P. falciparum sporozoite invasion. Additionally, it establishes a simple in vitro system to study the LS with improved invasion efficiency. This work paves the way for the greater malaria and liver biology communities to explore fundamental questions of hepatocyte-pathogen interactions and extend the system to other human malaria parasite species, like P. vivax.

Blocking of an intronic splicing silencer completely rescues IKBKAP exon 20 splicing in familial dysautonomia patient cells.

Familial dysautonomia (FD) is a severe genetic disorder causing sensory and autonomic dysfunction. It is predominantly caused by a c.2204+6T>C mutation in the IKBKAP gene. This mutation decreases the 5' splice site strength of IKBKAP exon 20 leading to exon 20 skipping and decreased amounts of full-length IKAP protein. We identified a binding site for the splicing regulatory protein hnRNP A1 downstream of the IKBKAP exon 20 5'-splice site. We show that hnRNP A1 binds to this splicing regulatory element (SRE) and that two previously described inhibitory SREs inside IKBKAP exon 20 are also bound by hnRNP A1. Knockdown of hnRNP A1 in FD patient fibroblasts increases IKBKAP exon 20 inclusion demonstrating that hnRNP A1 is a negative regulator of IKBKAP exon 20 splicing. Furthermore, by mutating the SREs in an IKBKAP minigene we show that all three SREs cause hnRNP A1-mediated exon repression. We designed splice switching oligonucleotides (SSO) that blocks the intronic hnRNP A1 binding site, and demonstrate that this completely rescues splicing of IKBKAP exon 20 in FD patient fibroblasts and increases the amounts of IKAP protein. We propose that this may be developed into a potential new specific treatment of FD.