Parkinson's disease and schizophrenia interactomes contain temporally distinct gene clusters underlying comorbid mechanisms and unique disease processes

Genome-wide association studies suggest significant overlaps in Parkinson's disease (PD) and schizophrenia (SZ) risks, but the underlying mechanisms remain elusive. The protein-protein interaction network ('interactome') plays a crucial role in PD and SZ and can incorporate their spatiotemporal specificities. Therefore, to study the linked biology of PD and SZ, we compiled PD- and SZ-associated genes from the DisGeNET database, and constructed their interactomes using BioGRID and HPRD. We examined the interactomes using clustering and enrichment analyses, in conjunction with the transcriptomic data of 26 brain regions spanning foetal stages to adulthood available in the BrainSpan Atlas. PD and SZ interactomes formed four gene clusters with distinct temporal identities (Disease Gene Networks or 'DGNs'1-4). DGN1 had unique SZ interactome genes highly expressed across developmental stages, corresponding to a neurodevelopmental SZ subtype. DGN2, containing unique SZ interactome genes expressed from early infancy to adulthood, correlated with an inflammation-driven SZ subtype and adult SZ risk. DGN3 contained unique PD interactome genes expressed in late infancy, early and late childhood, and adulthood, and involved in mitochondrial pathways. DGN4, containing prenatally-expressed genes common to both the interactomes, involved in stem cell pluripotency and overlapping with the interactome of 22q11 deletion syndrome (comorbid psychosis and Parkinsonism), potentially regulates neurodevelopmental mechanisms in PD-SZ comorbidity. Our findings suggest that disrupted neurodevelopment (regulated by DGN4) could expose risk windows in PD and SZ, later elevating disease risk through inflammation (DGN2). Alternatively, variant clustering in DGNs may produce disease subtypes, e.g., PD-SZ comorbidity with DGN4, and early/late-onset SZ with DGN1/DGN2.

Flow blockage disrupts cilia-driven fluid transport in the epileptic brain

A carpet of ependymal motile cilia lines the brain ventricular system, forming a network of flow channels and barriers that pattern cerebrospinal fluid (CSF) flow at the surface. This CSF transport system is evolutionary conserved, but its physiological function remains unknown. Here we investigated its potential role in epilepsy with studies focused on CDKL5 deficiency disorder (CDD), a neurodevelopmental disorder with early-onset epilepsy refractory to seizure medications and the most common cause of infant epilepsy. CDKL5 is a highly conserved X-linked gene suggesting its function in regulating cilia length and motion in the green alga Chlamydomonas might have implication in the etiology of CDD. Examination of the structure and function of airway motile cilia revealed both the CDD patients and the Cdkl5 knockout mice exhibit cilia lengthening and abnormal cilia motion. Similar defects were observed for brain ventricular cilia in the Cdkl5 knockout mice. Mapping ependymal cilia generated flow in the ventral third ventricle (v3V), a brain region with important physiological functions showed altered patterning of flow. Tracing of cilia-mediated inflow into v3V with fluorescent dye revealed the appearance of a flow barrier at the inlet of v3V in Cdkl5 knockout mice. Analysis of mice with a mutation in another epilepsy-associated kinase, Yes1, showed the same disturbance of cilia motion and flow patterning. The flow barrier was also observed in the Foxj1^± and FOXJ1Cre^ERT:Cdkl5^y/fl mice, confirming the contribution of ventricular cilia to the flow disturbances. Importantly, mice exhibiting altered cilia-driven flow also showed increased susceptibility to anesthesia-induced seizure-like activity. The cilia-driven flow disturbance arises from altered cilia beating orientation with the disrupted polarity of the cilia anchoring rootlet meshwork. Together these findings indicate motile cilia disturbances have an essential role in CDD-associated seizures and beyond, suggesting cilia regulating kinases may be a therapeutic target for medication-resistant epilepsy.

Uncompensated mitochondrial oxidative stress underlies heart failure in an iPSC-derived model of congenital heart disease

Hypoplastic left heart syndrome (HLHS) is a severe congenital heart disease with 30% mortality from heart failure (HF) in the first year of life, but the cause of early HF remains unknown. Induced pluripotent stem-cell-derived cardiomyocytes (iPSC-CM) from patients with HLHS showed that early HF is associated with increased apoptosis, mitochondrial respiration defects, and redox stress from abnormal mitochondrial permeability transition pore (mPTP) opening and failed antioxidant response. In contrast, iPSC-CM from patients without early HF showed normal respiration with elevated antioxidant response. Single-cell transcriptomics confirmed that early HF is associated with mitochondrial dysfunction accompanied with endoplasmic reticulum (ER) stress. These findings indicate that uncompensated oxidative stress underlies early HF in HLHS. Importantly, mitochondrial respiration defects, oxidative stress, and apoptosis were rescued by treatment with sildenafil to inhibit mPTP opening or TUDCA to suppress ER stress. Together these findings point to the potential use of patient iPSC-CM for modeling clinical heart failure and the development of therapeutics.

Novel Protein-Protein Interactions Highlighting the Crosstalk between Hypoplastic Left Heart Syndrome, Ciliopathies and Neurodevelopmental Delays

Hypoplastic left heart syndrome (HLHS) is a severe congenital heart disease (CHD) affecting 1 in 5000 newborns. We constructed the interactome of 74 HLHS-associated genes identified from a large-scale mouse mutagenesis screen, augmenting it with 408 novel protein-protein interactions (PPIs) using our High-Precision Protein-Protein Interaction Prediction (HiPPIP) model. The interactome is available on a webserver with advanced search capabilities. A total of 364 genes including 73 novel interactors were differentially regulated in tissue/iPSC-derived cardiomyocytes of HLHS patients. Novel PPIs facilitated the identification of TOR signaling and endoplasmic reticulum stress modules. We found that 60.5% of the interactome consisted of housekeeping genes that may harbor large-effect mutations and drive HLHS etiology but show limited transmission. Network proximity of diabetes, Alzheimer's disease, and liver carcinoma-associated genes to HLHS genes suggested a mechanistic basis for their comorbidity with HLHS. Interactome genes showed tissue-specificity for sites of extracardiac anomalies (placenta, liver and brain). The HLHS interactome shared significant overlaps with the interactomes of ciliopathy- and microcephaly-associated genes, with the shared genes enriched for genes involved in intellectual disability and/or developmental delay, and neuronal death pathways, respectively. This supported the increased burden of ciliopathy variants and prevalence of neurological abnormalities observed among HLHS patients with developmental delay and microcephaly, respectively.

Benchmark Evaluation of Protein-Protein Interaction Prediction Algorithms

Protein-protein interactions (PPIs) perform various functions and regulate processes throughout cells. Knowledge of the full network of PPIs is vital to biomedical research, but most of the PPIs are still unknown. As it is infeasible to discover all of them experimentally due to technical and resource limitations, computational prediction of PPIs is essential and accurately assessing the performance of algorithms is required before further application or translation. However, many published methods compose their evaluation datasets incorrectly, using a higher proportion of positive class data than occuring naturally, leading to exaggerated performance. We re-implemented various published algorithms and evaluated them on datasets with realistic data compositions and found that their performance is overstated in original publications; with several methods outperformed by our control models built on 'illogical' and random number features. We conclude that these methods are influenced by an over-characterization of some proteins in the literature and due to scale-free nature of PPI network and that they fail when tested on all possible protein pairs. Additionally, we found that sequence-only-based algorithms performed worse than those that employ functional and expression features. We present a benchmark evaluation of many published algorithms for PPI prediction. The source code of our implementations and the benchmark datasets created here are made available in open source.

Generalized and social anxiety disorder interactomes show distinctive overlaps with striosome and matrix interactomes

Mechanisms underlying anxiety disorders remain elusive despite the discovery of several associated genes. We constructed the protein-protein interaction networks (interactomes) of six anxiety disorders and noted enrichment for striatal expression among common genes in the interactomes. Five of these interactomes shared distinctive overlaps with the interactomes of genes that were differentially expressed in two striatal compartments (striosomes and matrix). Generalized anxiety disorder and social anxiety disorder interactomes showed exclusive and statistically significant overlaps with the striosome and matrix interactomes, respectively. Systematic gene expression analysis with the anxiety disorder interactomes constrained to contain only those genes that were shared with striatal compartment interactomes revealed a bifurcation among the disorders, which was influenced by the anterior cingulate cortex, nucleus accumbens, amygdala and hippocampus, and the dopaminergic signaling pathway. Our results indicate that the functionally distinct striatal pathways constituted by the striosome and the matrix may influence the etiological differentiation of various anxiety disorders.

Malignant Pleural Mesothelioma Interactome with 364 Novel Protein-Protein Interactions

Malignant pleural mesothelioma (MPM) is an aggressive cancer affecting the outer lining of the lung, with a median survival of less than one year. We constructed an 'MPM interactome' with over 300 computationally predicted protein-protein interactions (PPIs) and over 2400 known PPIs of 62 literature-curated genes whose activity affects MPM. Known PPIs of the 62 MPM associated genes were derived from Biological General Repository for Interaction Datasets (BioGRID) and Human Protein Reference Database (HPRD). Novel PPIs were predicted by applying the HiPPIP algorithm, which computes features of protein pairs such as cellular localization, molecular function, biological process membership, genomic location of the gene, and gene expression in microarray experiments, and classifies the pairwise features as interacting or non-interacting based on a random forest model. We validated five novel predicted PPIs experimentally. The interactome is significantly enriched with genes differentially ex-pressed in MPM tumors compared with normal pleura and with other thoracic tumors, genes whose high expression has been correlated with unfavorable prognosis in lung cancer, genes differentially expressed on crocidolite exposure, and exosome-derived proteins identified from malignant mesothelioma cell lines. 28 of the interactors of MPM proteins are targets of 147 U.S. Food and Drug Administration (FDA)-approved drugs. By comparing disease-associated versus drug-induced differential expression profiles, we identified five potentially repurposable drugs, namely cabazitaxel, primaquine, pyrimethamine, trimethoprim and gliclazide. Preclinical studies may be con-ducted in vitro to validate these computational results. Interactome analysis of disease-associated genes is a powerful approach with high translational impact. It shows how MPM-associated genes identified by various high throughput studies are functionally linked, leading to clinically translatable results such as repurposed drugs. The PPIs are made available on a webserver with interactive user interface, visualization and advanced search capabilities.

Cilia interactome with predicted protein-protein interactions reveals connections to Alzheimer's disease, aging and other neuropsychiatric processes

Cilia are dynamic microtubule-based organelles present on the surface of many eukaryotic cell types and can be motile or non-motile primary cilia. Cilia defects underlie a growing list of human disorders, collectively called ciliopathies, with overlapping phenotypes such as developmental delays and cognitive and memory deficits. Consistent with this, cilia play an important role in brain development, particularly in neurogenesis and neuronal migration. These findings suggest that a deeper systems-level understanding of how ciliary proteins function together may provide new mechanistic insights into the molecular etiologies of nervous system defects. Towards this end, we performed a protein-protein interaction (PPI) network analysis of known intraflagellar transport, BBSome, transition zone, ciliary membrane and motile cilia proteins. Known PPIs of ciliary proteins were assembled from online databases. Novel PPIs were predicted for each ciliary protein using a computational method we developed, called High-precision PPI Prediction (HiPPIP) model. The resulting cilia "interactome" consists of 165 ciliary proteins, 1,011 known PPIs, and 765 novel PPIs. The cilia interactome revealed interconnections between ciliary proteins, and their relation to several pathways related to neuropsychiatric processes, and to drug targets. Approximately 184 genes in the cilia interactome are targeted by 548 currently approved drugs, of which 103 are used to treat various diseases of nervous system origin. Taken together, the cilia interactome presented here provides novel insights into the relationship between ciliary protein dysfunction and neuropsychiatric disorders, for e.g. interconnections of Alzheimer's disease, aging and cilia genes. These results provide the framework for the rational design of new therapeutic agents for treatment of ciliopathies and neuropsychiatric disorders.

Interactome of SARS-CoV-2 / nCoV19 modulated host proteins with computationally predicted PPIs

World over, people are looking for solutions to tackle the pandemic coronavirus disease (COVID-19) caused by the virus SARS-CoV-2/nCoV-19. Notable contributions in biomedical field have been characterizing viral genomes, host transcriptomes and proteomes, repurposable drugs and vaccines. In one such study, 332 human proteins targeted by nCoV19 were identified. We expanded this set of host proteins by constructing their protein interactome, including in it not only the known protein-protein interactions (PPIs) but also novel, hitherto unknown PPIs predicted with our High-precision Protein-Protein Interaction Prediction (HiPPIP) model that was shown to be highly accurate. In fact, one of the earliest discoveries made possible by HiPPIP is related to activation of immunity upon viral infection. We found that several interactors of the host proteins are differentially expressed upon viral infection, are related to highly relevant pathways, and that the novel interaction of NUP98 with CHMP5 may activate an antiviral mechanism leading to disruption of viral budding. We are making the interactions available as downloadable files to facilitate future systems biology studies and also on a web-server at http://hagrid.dbmi.pitt.edu/corona that allows not only keyword search but also queries such as "PPIs where one protein is associated with 'virus' and the interactors with 'pulmonary'".

Potentially repurposable drugs for schizophrenia identified from its interactome

We previously presented the protein-protein interaction network of schizophrenia associated genes, and from it, the drug-protein interactome which showed the drugs that target any of the proteins in the interactome. Here, we studied these drugs further to identify whether any of them may potentially be repurposable for schizophrenia. In schizophrenia, gene expression has been described as a measurable aspect of the disease reflecting the action of risk genes. We studied each of the drugs from the interactome using the BaseSpace Correlation Engine, and shortlisted those that had a negative correlation with differential gene expression of schizophrenia. This analysis resulted in 12 drugs whose differential gene expression (drug versus normal) had an anti-correlation with differential expression for schizophrenia (disorder versus normal). Some of these drugs were already being tested for their clinical activity in schizophrenia and other neuropsychiatric disorders. Several proteins in the protein interactome of the targets of several of these drugs were associated with various neuropsychiatric disorders. The network of genes with opposite drug-induced versus schizophrenia-associated expression profiles were significantly enriched in pathways relevant to schizophrenia etiology and GWAS genes associated with traits or diseases that had a pathophysiological overlap with schizophrenia. Drugs that targeted the same genes as the shortlisted drugs, have also demonstrated clinical activity in schizophrenia and other related disorders. This integrated computational analysis will help translate insights from the schizophrenia drug-protein interactome to clinical research - an important step, especially in the field of psychiatric drug development which faces a high failure rate.

Towards Extracting Supporting Information About Predicted Protein-Protein Interactions

One of the goals of relation extraction is to identify protein-protein interactions (PPIs) in biomedical literature. Current systems are capturing binary relations and also the direction and type of an interaction. Besides assisting in the curation PPIs into databases, there has been little real-world application of these algorithms. We describe UPSITE, a text mining tool for extracting evidence in support of a hypothesized interaction. Given a predicted PPI, UPSITE uses a binary relation detector to check whether a PPI is found in abstracts in PubMed. If it is not found, UPSITE retrieves documents relevant to each of the two proteins separately, and extracts contextual information about biological events surrounding each protein, and calculates semantic similarity of the two proteins to provide evidential support for the predicted PPI. In evaluations, relation extraction achieved an Fscore of 0.88 on the HPRD50 corpus, and semantic similarity measured with angular distance was found to be statistically significant. With the development of PPI prediction algorithms, the burden of interpreting the validity and relevance of novel PPIs is on biologists. We suggest that presenting annotations of the two proteins in a PPI side-by-side and a score that quantifies their similarity lessens this burden to some extent.

Is the average shortest path length of gene set a reflection of their biological relatedness?

When a set of genes are identified to be related to a disease, say through gene expression analysis, it is common to examine the average distance among their protein products in the human interactome as a measure of biological relatedness of these genes. The reasoning for this is that, genes associated with a disease would tend to be functionally related, and that functionally related genes would be closely connected to each other in the interactome. Typically, average shortest path length (ASPL) of disease genes (although referred to as genes in the context of disease-associations, the interactions are among protein-products of these genes) is compared to ASPL of randomly selected genes or to ASPL in a randomly permuted network. We examined whether the ASPL of a set of genes is indeed a good measure of biological relatedness or whether it is simply a characteristic of the degree distribution of those genes. We examined the ASPL of genes sets of some disease and pathway associations and compared them to ASPL of three types of randomly selected control sets: uniform selection, from entire proteome, degree-matched selection, and random permutation of the network. We found that disease associated genes and their degree-matched random genes have comparable ASPL. In other words, ASPL is a characteristic of the degree of the genes and the network topology, and not that of functional coherence.

Generating testable hypotheses for schizophrenia and rheumatoid arthritis pathogenesis by integrating epidemiological, genomic, and protein interaction data

Patients with schizophrenia and their relatives have reduced prevalence of rheumatoid arthritis. Schizophrenia and rheumatoid arthritis genome-wide association studies also indicate negative genetic correlations, suggesting that there may be shared pathogenesis at the DNA level or downstream. A portion of the inverse prevalence could be attributed to pleiotropy, i.e., variants of a single nucleotide polymorphism that could confer differential risk for these disorders. To study the basis for such an interrelationship, we initially compared lists of single nucleotide polymorphisms with significant genetic associations (p < 1e-8) for schizophrenia or rheumatoid arthritis, evaluating patterns of linkage disequilibrium and apparent pleiotropic risk profiles. Single nucleotide polymorphisms that conferred risk for both schizophrenia and rheumatoid arthritis were localized solely to the extended HLA region. Among single nucleotide polymorphisms that conferred differential risk for schizophrenia and rheumatoid arthritis, the majority were localized to HLA-B, TNXB, NOTCH4, HLA-C, HCP5, MICB, PSORS1C1, and C6orf10; published functional data indicate that HLA-B and HLA-C have the most plausible pathogenic roles in both disorders. Interactomes of these eight genes were constructed from protein-protein interaction information using publicly available databases and novel computational predictions. The genes harboring apparently pleiotropic single nucleotide polymorphisms are closely connected to rheumatoid arthritis and schizophrenia associated genes through common interacting partners. A separate and independent analysis of the interactomes of rheumatoid arthritis and schizophrenia genes showed a significant overlap between the two interactomes and that they share several common pathways, motivating functional studies suggesting a relationship in the pathogenesis of schizophrenia/rheumatoid arthritis.

A pilot study on the prevalence of DNA palindromes in breast cancer genomes

Background

DNA palindromes are a unique pattern of repeat sequences that are present in the human genome. It consists of a sequence of nucleotides in which the second half is the complement of the first half but appearing in reverse order. These palindromic sequences may have a significant role in DNA replication, transcription and gene regulation processes. They occur frequently in human cancers by clustering at specific locations of the genome that undergo gene amplification and tumorigenesis. Moreover, some studies showed that palindromes are clustered in amplified regions of breast cancer genomes especially in chromosomes (chr) 8 and 11. With the large number of personal genomes and cancer genomes becoming available, it is now possible to study their association to diseases using computational methods. Here, we conducted a pilot study on chromosomes 8 and 11 of cancer genomes to identify computationally the differentially occurring palindromes.

Methods

We processed 69 breast cancer genomes from The Cancer Genome Atlas including serum-normal and tumor genomes, and 1000 Genomes to serve as control group. The Biological Language Modelling Toolkit (BLMT) computes palindromes in whole genomes. We developed a computational pipeline integrating BLMT to compute and compare prevalence of palindromes in personal genomes.

Results

We carried out a pilot study on chr 8 and chr 11 taking into account single nucleotide polymorphisms, insertions and deletions. Of all the palindromes that showed any variation in cancer genomes, 38% of what were near breast cancer genes happened to be the most differentiated palindromes in tumor (i.e. they ranked among the top 25% by our heuristic measure).

Conclusions

These results will shed light on the prevalence of palindromes in oncogenes and the mutations that are present in the palindromic regions that could contribute to genomic rearrangements, and breast cancer progression.

Predicted protein interactions of IFITMs may shed light on mechanisms of Zika virus-induced microcephaly and host invasion

After the first reported case of Zika virus (ZIKV) in Brazil, in 2015, a significant increase in the reported cases of microcephaly was observed. Microcephaly is a neurological condition in which the infant’s head is significantly smaller with complications in brain development. Recently, two small membrane-associated interferon-inducible transmembrane proteins (IFITM1 and IFITM3) have been shown to repress members of the flaviviridae family which includes ZIKV. However, the exact mechanisms leading to the inhibition of the virus are yet unknown. Here, we assembled an interactome of IFITM1 and IFITM3 with known protein-protein interactions (PPIs) collected from publicly available databases and novel PPIs predicted using the High-confidence Protein-Protein Interaction Prediction (HiPPIP) model. We analyzed the functional and pathway associations of the interacting proteins, and found that there are several immunity pathways (toll-like receptor signaling, cd28 signaling in T-helper cells, crosstalk between dendritic cells and natural killer cells), neuronal pathways (axonal guidance signaling, neural tube closure and actin cytoskeleton signaling) and developmental pathways (neural tube closure, embryonic skeletal system development) that are associated with these interactors. Our novel PPIs associate cilia dysfunction in ependymal cells to microcephaly, and may also shed light on potential targets of ZIKV for host invasion by immunosuppression and cytoskeletal rearrangements. These results could help direct future research in elucidating the mechanisms underlying host defense to ZIKV and other flaviviruses.

LENS: web-based lens for enrichment and network studies of human proteins

Background

Network analysis is a common approach for the study of genetic view of diseases and biological pathways. Typically, when a set of genes are identified to be of interest in relation to a disease, say through a genome wide association study (GWAS) or a different gene expression study, these genes are typically analyzed in the context of their protein-protein interaction (PPI) networks. Further analysis is carried out to compute the enrichment of known pathways and disease-associations in the network. Having tools for such analysis at the fingertips of biologists without the requirement for computer programming or curation of data would accelerate the characterization of genes of interest. Currently available tools do not integrate network and enrichment analysis and their visualizations, and most of them present results in formats not most conducive to human cognition.

Results

We developed the tool Lens for Enrichment and Network Studies of human proteins (LENS) that performs network and pathway and diseases enrichment analyses on genes of interest to users. The tool creates a visualization of the network, provides easy to read statistics on network connectivity, and displays Venn diagrams with statistical significance values of the network's association with drugs, diseases, pathways, and GWASs. We used the tool to analyze gene sets related to craniofacial development, autism, and schizophrenia.

Conclusion

LENS is a web-based tool that does not require and download or plugins to use. The tool is free and does not require login for use, and is available at http://severus.dbmi.pitt.edu/LENS.

Global genetic analysis in mice unveils central role for cilia in congenital heart disease

Congenital heart disease (CHD) is the most prevalent birth defect, affecting nearly 1% of live births; the incidence of CHD is up to tenfold higher in human fetuses. A genetic contribution is strongly suggested by the association of CHD with chromosome abnormalities and high recurrence risk. Here we report findings from a recessive forward genetic screen in fetal mice, showing that cilia and cilia-transduced cell signalling have important roles in the pathogenesis of CHD. The cilium is an evolutionarily conserved organelle projecting from the cell surface with essential roles in diverse cellular processes. Using echocardiography, we ultrasound scanned 87,355 chemically mutagenized C57BL/6J fetal mice and recovered 218 CHD mouse models. Whole-exome sequencing identified 91 recessive CHD mutations in 61 genes. This included 34 cilia-related genes, 16 genes involved in cilia-transduced cell signalling, and 10 genes regulating vesicular trafficking, a pathway important for ciliogenesis and cell signalling. Surprisingly, many CHD genes encoded interacting proteins, suggesting that an interactome protein network may provide a larger genomic context for CHD pathogenesis. These findings provide novel insights into the potential Mendelian genetic contribution to CHD in the fetal population, a segment of the human population not well studied. We note that the pathways identified show overlap with CHD candidate genes recovered in CHD patients, suggesting that they may have relevance to the more complex genetics of CHD overall. These CHD mouse models and >8,000 incidental mutations have been sperm archived, creating a rich public resource for human disease modelling.

Antiviral activity of human OASL protein is mediated by enhancing signaling of the RIG-I RNA sensor

Virus infection is sensed in the cytoplasm by retinoic acid-inducible gene I (RIG-I, also known as DDX58), which requires RNA and polyubiquitin binding to induce type I interferon (IFN) and activate cellular innate immunity. We show that the human IFN-inducible oligoadenylate synthetases-like (OASL) protein has antiviral activity and mediates RIG-I activation by mimicking polyubiquitin. Loss of OASL expression reduced RIG-I signaling and enhanced virus replication in human cells. Conversely, OASL expression suppressed replication of a number of viruses in a RIG-I-dependent manner and enhanced RIG-I-mediated IFN induction. OASL interacted and colocalized with RIG-I, and through its C-terminal ubiquitin-like domain specifically enhanced RIG-I signaling. Bone-marrow-derived macrophages from mice deficient for Oasl2 showed that among the two mouse orthologs of human OASL, Oasl2 is functionally similar to human OASL. Our findings show a mechanism by which human OASL contributes to host antiviral responses by enhancing RIG-I activation.

Predicting gene ontology annotations of orphan GWAS genes using protein-protein interactions

Background

The number of genome-wide association studies (GWAS) has increased rapidly in the past couple of years, resulting in the identification of genes associated with different diseases. The next step in translating these findings into biomedically useful information is to find out the mechanism of the action of these genes. However, GWAS studies often implicate genes whose functions are currently unknown; for example, MYEOV, ANKLE1, TMEM45B and ORAOV1 are found to be associated with breast cancer, but their molecular function is unknown.

Results

We carried out Bayesian inference of Gene Ontology (GO) term annotations of genes by employing the directed acyclic graph structure of GO and the network of protein-protein interactions (PPIs). The approach is designed based on the fact that two proteins that interact biophysically would be in physical proximity of each other, would possess complementary molecular function, and play role in related biological processes. Predicted GO terms were ranked according to their relative association scores and the approach was evaluated quantitatively by plotting the precision versus recall values and F-scores (the harmonic mean of precision and recall) versus varying thresholds. Precisions of ~58% and ~ 40% for localization and functions respectively of proteins were determined at a threshold of ~30 (top 30 GO terms in the ranked list). Comparison with function prediction based on semantic similarity among nodes in an ontology and incorporation of those similarities in a k-nearest neighbor classifier confirmed that our results compared favorably.

Conclusions

This approach was applied to predict the cellular component and molecular function GO terms of all human proteins that have interacting partners possessing at least one known GO annotation. The list of predictions is available at http://severus.dbmi.pitt.edu/engo/GOPRED.html. We present the algorithm, evaluations and the results of the computational predictions, especially for genes identified in GWAS studies to be associated with diseases, which are of translational interest.

Research prioritization through prediction of future impact on biomedical science: a position paper on inference-analytics

Background

Advances in biotechnology have created “big-data” situations in molecular and cellular biology. Several sophisticated algorithms have been developed that process big data to generate hundreds of biomedical hypotheses (or predictions). The bottleneck to translating this large number of biological hypotheses is that each of them needs to be studied by experimentation for interpreting its functional significance. Even when the predictions are estimated to be very accurate, from a biologist’s perspective, the choice of which of these predictions is to be studied further is made based on factors like availability of reagents and resources and the possibility of formulating some reasonable hypothesis about its biological relevance. When viewed from a global perspective, say from that of a federal funding agency, ideally the choice of which prediction should be studied would be made based on which of them can make the most translational impact.

Results

We propose that algorithms be developed to identify which of the computationally generated hypotheses have potential for high translational impact; this way, funding agencies and scientific community can invest resources and drive the research based on a global view of biomedical impact without being deterred by local view of feasibility. In short, data-analytic algorithms analyze big-data and generate hypotheses; in contrast, the proposed inference-analytic algorithms analyze these hypotheses and rank them by predicted biological impact. We demonstrate this through the development of an algorithm to predict biomedical impact of protein-protein interactions (PPIs) which is estimated by the number of future publications that cite the paper which originally reported the PPI.

Conclusions

This position paper describes a new computational problem that is relevant in the era of big-data and discusses the challenges that exist in studying this problem, highlighting the need for the scientific community to engage in this line of research. The proposed class of algorithms, namely inference-analytic algorithms, is necessary to ensure that resources are invested in translating those computational outcomes that promise maximum biological impact. Application of this concept to predict biomedical impact of PPIs illustrates not only the concept, but also the challenges in designing these algorithms.

Wiki-pi: a web-server of annotated human protein-protein interactions to aid in discovery of protein function

Protein-protein interactions (PPIs) are the basis of biological functions. Knowledge of the interactions of a protein can help understand its molecular function and its association with different biological processes and pathways. Several publicly available databases provide comprehensive information about individual proteins, such as their sequence, structure, and function. There also exist databases that are built exclusively to provide PPIs by curating them from published literature. The information provided in these web resources is protein-centric, and not PPI-centric. The PPIs are typically provided as lists of interactions of a given gene with links to interacting partners; they do not present a comprehensive view of the nature of both the proteins involved in the interactions. A web database that allows search and retrieval based on biomedical characteristics of PPIs is lacking, and is needed. We present Wiki-Pi (read Wiki-π), a web-based interface to a database of human PPIs, which allows users to retrieve interactions by their biomedical attributes such as their association to diseases, pathways, drugs and biological functions. Each retrieved PPI is shown with annotations of both of the participant proteins side-by-side, creating a basis to hypothesize the biological function facilitated by the interaction. Conceptually, it is a search engine for PPIs analogous to PubMed for scientific literature. Its usefulness in generating novel scientific hypotheses is demonstrated through the study of IGSF21, a little-known gene that was recently identified to be associated with diabetic retinopathy. Using Wiki-Pi, we infer that its association to diabetic retinopathy may be mediated through its interactions with the genes HSPB1, KRAS, TMSB4X and DGKD, and that it may be involved in cellular response to external stimuli, cytoskeletal organization and regulation of molecular activity. The website also provides a wiki-like capability allowing users to describe or discuss an interaction. Wiki-Pi is available publicly and freely at http://severus.dbmi.pitt.edu/wiki-pi/.

An efficient heuristic method for active feature acquisition and its application to protein-protein interaction prediction

BACKGROUND

Machine learning approaches for classification learn the pattern of the feature space of different classes, or learn a boundary that separates the feature space into different classes. The features of the data instances are usually available, and it is only the class-labels of the instances that are unavailable. For example, to classify text documents into different topic categories, the words in the documents are features and they are readily available, whereas the topic is what is predicted. However, in some domains obtaining features may be resource-intensive because of which not all features may be available. An example is that of protein-protein interaction prediction, where not only are the labels ('interacting' or 'non-interacting') unavailable, but so are some of the features. It may be possible to obtain at least some of the missing features by carrying out a few experiments as permitted by the available resources. If only a few experiments can be carried out to acquire missing features, which proteins should be studied and which features of those proteins should be determined? From the perspective of machine learning for PPI prediction, it would be desirable that those features be acquired which when used in training the classifier, the accuracy of the classifier is improved the most. That is, the utility of the feature-acquisition is measured in terms of how much acquired features contribute to improving the accuracy of the classifier. Active feature acquisition (AFA) is a strategy to preselect such instance-feature combinations (i.e. protein and experiment combinations) for maximum utility. The goal of AFA is the creation of optimal training set that would result in the best classifier, and not in determining the best classification model itself.

RESULTS

We present a heuristic method for active feature acquisition to calculate the utility of acquiring a missing feature. This heuristic takes into account the change in belief of the classification model induced by the acquisition of the feature under consideration. As compared to random selection of proteins on which the experiments are performed and the type of experiment that is performed, the heuristic method reduces the number of experiments to as few as 40%. Most notable characteristic of this method is that it does not require re-training of the classification model on every possible combination of instance, feature and feature-value tuples. For this reason, our method is far less computationally expensive as compared with previous AFA strategies.

CONCLUSIONS

The results show that our heuristic method for AFA creates an optimal training set with far less features acquired as compared to random acquisition. This shows the value of active feature acquisition to aid in protein-protein interaction prediction where feature acquisition is costly. Compared to previous methods, the proposed method reduces computational cost while also achieving a better F-score. The proposed method is valuable as it presents a direction to AFA with a far lesser computational expense by removing the need for the first time, of training a classifier for every combination of instance, feature and feature-value tuples which would be impractical for several domains.

Suite of tools for statistical N-gram language modeling for pattern mining in whole genome sequences

Genome sequences contain a number of patterns that have biomedical significance. Repetitive sequences of various kinds are a primary component of most of the genomic sequence patterns. We extended the suffix-array based Biological Language Modeling Toolkit to compute n-gram frequencies as well as n-gram language-model based perplexity in windows over the whole genome sequence to find biologically relevant patterns. We present the suite of tools and their application for analysis on whole human genome sequence.

N-gram analysis of 970 microbial organisms reveals presence of biological language models

Background

It has been suggested previously that genome and proteome sequences show characteristics typical of natural-language texts such as "signature-style" word usage indicative of authors or topics, and that the algorithms originally developed for natural language processing may therefore be applied to genome sequences to draw biologically relevant conclusions. Following this approach of 'biological language modeling', statistical n-gram analysis has been applied for comparative analysis of whole proteome sequences of 44 organisms. It has been shown that a few particular amino acid n-grams are found in abundance in one organism but occurring very rarely in other organisms, thereby serving as genome signatures. At that time proteomes of only 44 organisms were available, thereby limiting the generalization of this hypothesis. Today nearly 1,000 genome sequences and corresponding translated sequences are available, making it feasible to test the existence of biological language models over the evolutionary tree.

Results

We studied whole proteome sequences of 970 microbial organisms using n-gram frequencies and cross-perplexity employing the Biological Language Modeling Toolkit and Patternix Revelio toolkit. Genus-specific signatures were observed even in a simple unigram distribution. By taking statistical n-gram model of one organism as reference and computing cross-perplexity of all other microbial proteomes with it, cross-perplexity was found to be predictive of branch distance of the phylogenetic tree. For example, a 4-gram model from proteome of Shigellae flexneri 2a, which belongs to the Gammaproteobacteria class showed a self-perplexity of 15.34 while the cross-perplexity of other organisms was in the range of 15.59 to 29.5 and was proportional to their branching distance in the evolutionary tree from S. flexneri. The organisms of this genus, which happen to be pathotypes of E.coli, also have the closest perplexity values with E. coli.

Conclusion

Whole proteome sequences of microbial organisms have been shown to contain particular n-gram sequences in abundance in one organism but occurring very rarely in other organisms, thereby serving as proteome signatures. Further it has also been shown that perplexity, a statistical measure of similarity of n-gram composition, can be used to predict evolutionary distance within a genus in the phylogenetic tree.

Active learning for human protein-protein interaction prediction

BACKGROUND

Biological processes in cells are carried out by means of protein-protein interactions. Determining whether a pair of proteins interacts by wet-lab experiments is resource-intensive; only about 38,000 interactions, out of a few hundred thousand expected interactions, are known today. Active machine learning can guide the selection of pairs of proteins for future experimental characterization in order to accelerate accurate prediction of the human protein interactome.

RESULTS

Random forest (RF) has previously been shown to be effective for predicting protein-protein interactions. Here, four different active learning algorithms have been devised for selection of protein pairs to be used to train the RF. With labels of as few as 500 protein-pairs selected using any of the four active learning methods described here, the classifier achieved a higher F-score (harmonic mean of Precision and Recall) than with 3000 randomly chosen protein-pairs. F-score of predicted interactions is shown to increase by about 15% with active learning in comparison to that with random selection of data.

CONCLUSION

Active learning algorithms enable learning more accurate classifiers with much lesser labelled data and prove to be useful in applications where manual annotation of data is formidable. Active learning techniques demonstrated here can also be applied to other proteomics applications such as protein structure prediction and classification.

Active machine learning for transmembrane helix prediction

Background

About 30% of genes code for membrane proteins, which are involved in a wide variety of crucial biological functions. Despite their importance, experimentally determined structures correspond to only about 1.7% of protein structures deposited in the Protein Data Bank due to the difficulty in crystallizing membrane proteins. Algorithms that can identify proteins whose high-resolution structure can aid in predicting the structure of many previously unresolved proteins are therefore of potentially high value. Active machine learning is a supervised machine learning approach which is suitable for this domain where there are a large number of sequences but only very few have known corresponding structures. In essence, active learning seeks to identify proteins whose structure, if revealed experimentally, is maximally predictive of others.

Results

An active learning approach is presented for selection of a minimal set of proteins whose structures can aid in the determination of transmembrane helices for the remaining proteins. TMpro, an algorithm for high accuracy TM helix prediction we previously developed, is coupled with active learning. We show that with a well-designed selection procedure, high accuracy can be achieved with only few proteins. TMpro, trained with a single protein achieved an F-score of 94% on benchmark evaluation and 91% on MPtopo dataset, which correspond to the state-of-the-art accuracies on TM helix prediction that are achieved usually by training with over 100 training proteins.

Conclusion

Active learning is suitable for bioinformatics applications, where manually characterized data are not a comprehensive representation of all possible data, and in fact can be a very sparse subset thereof. It aids in selection of data instances which when characterized experimentally can improve the accuracy of computational characterization of remaining raw data. The results presented here also demonstrate that the feature extraction method of TMpro is well designed, achieving a very good separation between TM and non TM segments.