Search

Motivation Advances in RNA sequencing technologies have achieved an unprecedented accuracy in the quantification of mRNA isoforms, but our knowledge of isoform-specific functions has lagged behind. There is a need to understand the functional consequences of differential splicing, which could be supported by the generation of accurate and comprehensive isoform-specific gene ontology annotations. Results We present isoform interpretation, a method that uses expectation–maximization to infer isoform-specific functions based on the relationship between sequence and functional isoform similarity. We predicted isoform-specific functional annotations for 85 617 isoforms of 17 900 protein-coding human genes spanning a range of 17 430 distinct gene ontology terms. Comparison with a gold-standard corpus of manually annotated human isoform functions showed that isoform interpretation significantly outperforms state-of-the-art competing methods. We provide experimental evidence that functionally related isoforms predicted by isoform interpretation show a higher degree of domain sharing and expression correlation than functionally related genes. We also show that isoform sequence similarity correlates better with inferred isoform function than with gene-level function. Availability and implementation Source code, documentation, and resource files are freely available under a GNU3 license at https://github.com/TheJacksonLaboratory/isopretEM and https://zenodo.org/record/7594321.

Graph representation learning is a family of related approaches that learn low-dimensional vector representations of nodes and other graph elements called embeddings. Embeddings approximate characteristics of the graph and can be used for a variety of machine-learning tasks such as novel edge prediction. For many biomedical applications, partial knowledge exists about positive edges that represent relationships between pairs of entities, but little to no knowledge is available about negative edges that represent the explicit lack of a relationship between two nodes. For this reason, classification procedures are forced to assume that the vast majority of unlabeled edges are negative. Existing approaches to sampling negative edges for training and evaluating classifiers do so by uniformly sampling pairs of nodes. We show here that this sampling strategy typically leads to sets of positive and negative edges with imbalanced edge degree distributions. Using representative homogeneous and heterogeneous biomedical knowledge graphs, we show that this strategy artificially inflates measured classification performance. We present a degree-aware node sampling approach for sampling negative edge examples that mitigates this effect and is simple to implement.

BackgroundWith the continuing COVID-19 pandemic, identifying medications that improve COVID-19 outcomes is crucial. Studies suggest that use of metformin, an oral antihyperglycemic, is associated with reduced COVID-19 severity in individuals with diabetes compared to other antihyperglycemic medications. Some patients without diabetes, including those with polycystic ovary syndrome (PCOS) and prediabetes, are prescribed metformin for off-label use, which provides an opportunity to further investigate the effect of metformin on COVID-19.ParticipantsIn this observational, retrospective analysis, we leveraged the harmonized electronic health record data from 53 hospitals to construct cohorts of COVID-19 positive, metformin users without diabetes and propensity-weighted control users of levothyroxine (a medication for hypothyroidism that is not known to affect COVID-19 outcome) who had either PCOS (n = 282) or prediabetes (n = 3136). The primary outcome of interest was COVID-19 severity, which was classified as: mild, mild ED (emergency department), moderate, severe, or mortality/hospice.ResultsIn the prediabetes cohort, metformin use was associated with a lower rate of COVID-19 with severity of mild ED or worse (OR: 0.630, 95% CI 0.450 - 0.882, p < 0.05) and a lower rate of COVID-19 with severity of moderate or worse (OR: 0.490, 95% CI 0.336 - 0.715, p < 0.001). In patients with PCOS, we found no significant association between metformin use and COVID-19 severity, although the number of patients was relatively small.ConclusionsMetformin was associated with less severe COVID-19 in patients with prediabetes, as seen in previous studies of patients with diabetes. This is an important finding, since prediabetes affects between 19 and 38% of the US population, and COVID-19 is an ongoing public health emergency. Further observational and prospective studies will clarify the relationship between metformin and COVID-19 severity in patients with prediabetes, and whether metformin usage may reduce COVID-19 severity.

Graph Representation Learning methods opened new possibilities for addressing complex,real-world problems represented by graphs. However, many graphs used in these applicationscomprise millions of nodes and billions of edges and are beyond the capabilities of current methodsand software implementations. We present GRAPE, a software resource for graph processing andrepresentation learning that is able to scale with big graphs by using specialized and smart datastructures, algorithms, and a fast parallel implementation. When compared with state of the artsoftware resources, GRAPE shows an improvement of orders of magnitude in empirical space andtime complexity, as well as a substantial and statistically significant improvement in edge predictionand node label prediction performance. Furthermore, GRAPE provides over 80,000 graphs fromthe literature and other sources, standardized interfaces allowing a straightforward integration ofthird-party libraries, 61 node embedding methods, 25 inference models, and 3 modular pipelinesto allow a FAIR and reproducible comparison of methods and libraries for graph processing andembedding.

Stratification of patients with post-acute sequelae of SARS-CoV-2 infection (PASC, or long COVID) would allow precision clinical management strategies. However, long COVID is incompletely understood and characterised by a wide range of manifestations that are difficult to analyse computationally. Additionally, the generalisability of machine learning classification of COVID-19 clinical outcomes has rarely been tested.We present a method for computationally modelling PASC phenotype data based on electronic healthcare records (EHRs) and for assessing pairwise phenotypic similarity between patients using semantic similarity. Our approach defines a nonlinear similarity function that maps from a feature space of phenotypic abnormalities to a matrix of pairwise patient similarity that can be clustered using unsupervised machine learning.We found six clusters of PASC patients, each with distinct profiles of phenotypic abnormalities, including clusters with distinct pulmonary, neuropsychiatric, and cardiovascular abnormalities, and a cluster associated with broad, severe manifestations and increased mortality. There was significant association of cluster membership with a range of pre-existing conditions and measures of severity during acute COVID-19. We assigned new patients from other healthcare centres to clusters by maximum semantic similarity to the original patients, and showed that the clusters were generalisable across different hospital systems. The increased mortality rate originally identified in one cluster was consistently observed in patients assigned to that cluster in other hospital systems.Semantic phenotypic clustering provides a foundation for assigning patients to stratified subgroups for natural history or therapy studies on PASC.NIH (TR002306/OT2HL161847-01/OD011883/HG010860), U.S.D.O.E. (DE-AC02-05CH11231), Donald A. Roux Family Fund at Jackson Laboratory, Marsico Family at CU Anschutz.

Accurate stratification of patients with post-acute sequelae of SARS-CoV-2 infection (PASC, or long COVID) would allow precision clinical management strategies. However, the natural history of long COVID is incompletely understood and characterized by an extremely wide range of manifestations that are difficult to analyze computationally. In addition, the generalizability of machine learning classification of COVID-19 clinical outcomes has rarely been tested. We present a method for computationally modeling PASC phenotype data based on electronic healthcare records (EHRs) and for assessing pairwise phenotypic similarity between patients using semantic similarity. Our approach defines a nonlinear similarity function that maps from a feature space of phenotypic abnormalities to a matrix of pairwise patient similarity that can be clustered using unsupervised machine learning procedures. Using k-means clustering of this similarity matrix, we found six distinct clusters of PASC patients, each with distinct profiles of phenotypic abnormalities. There was a significant association of cluster membership with a range of pre-existing conditions and with measures of severity during acute COVID-19. Two of the clusters were associated with severe manifestations and displayed increased mortality. We assigned new patients from other healthcare centers to one of the six clusters on the basis of maximum semantic similarity to the original patients. We show that the identified clusters were generalizable across different hospital systems and that the increased mortality rate was consistently observed in two of the clusters. Semantic phenotypic clustering can provide a foundation for assigning patients to stratified subgroups for natural history or therapy studies on PASC.

Gene Ontology (GO) overrepresentation analysis characterizes the biological mechanisms common to sets of differentially expressed genes identified by high-throughput experiments. To date, GO overrepresentation analysis has mainly been used to evaluate differentially expressed genes, but short- and long-read RNA-seq technologies now allow increasingly accurate identification of differential alternative splicing. The function of most splice isoforms remain unknown, but if acccurate predictions could be made, overrepresentation analysis could be applied to differentially spliced isoforms to assess the functional implications of alternative splicing in RNA-seq experiments. We present isopret (Isoform Interpretation), a new paradigm for isoform function prediction based on the expectation-maximization framework. isopret leverages the relationships between sequence and functional isoform similarity to infer isoform specific functions in a highly accurate fashion. This enabled us to adapt GO overrepresentation analysis, which to date has been limited to differential gene expression, to be extended to assess overrepresentation of GO annotations in differentially spliced isoforms. An analysis of 100 RNA-seq studies including investigations of development, cancer, and common disease demonstrated that expression and splicing regulate different sets of biological functions. We make isopret predictions freely available in a desktop application that can be used to analyze differential expression and splicing in any bulk RNA-seq dataset.

Background Non-steroidal anti-inflammatory drugs (NSAIDs) are commonly used to reduce pain, fever, and inflammation but have been associated with complications in community-acquired pneumonia. Observations shortly after the start of the COVID-19 pandemic in 2020 suggested that ibuprofen was associated with an increased risk of adverse events in COVID-19 patients, but subsequent observational studies failed to demonstrate increased risk and in one case showed reduced risk associated with NSAID use. Methods A 38-center retrospective cohort study was performed that leveraged the harmonized, high-granularity electronic health record data of the National COVID Cohort Collaborative. A propensity-matched cohort of 19,746 COVID-19 inpatients was constructed by matching cases (treated with NSAIDs at the time of admission) and 19,746 controls (not treated) from 857,061 patients with COVID-19 available for analysis. The primary outcome of interest was COVID-19 severity in hospitalized patients, which was classified as: moderate, severe, or mortality/hospice. Secondary outcomes were acute kidney injury (AKI), extracorporeal membrane oxygenation (ECMO), invasive ventilation, and all-cause mortality at any time following COVID-19 diagnosis. Results Logistic regression showed that NSAID use was not associated with increased COVID-19 severity (OR: 0.57 95% CI: 0.53–0.61). Analysis of secondary outcomes using logistic regression showed that NSAID use was not associated with increased risk of all-cause mortality (OR 0.51 95% CI: 0.47–0.56), invasive ventilation (OR: 0.59 95% CI: 0.55–0.64), AKI (OR: 0.67 95% CI: 0.63–0.72), or ECMO (OR: 0.51 95% CI: 0.36–0.7). In contrast, the odds ratios indicate reduced risk of these outcomes, but our quantitative bias analysis showed E-values of between 1.9 and 3.3 for these associations, indicating that comparatively weak or moderate confounder associations could explain away the observed associations. Conclusions Study interpretation is limited by the observational design. Recording of NSAID use may have been incomplete. Our study demonstrates that NSAID use is not associated with increased COVID-19 severity, all-cause mortality, invasive ventilation, AKI, or ECMO in COVID-19 inpatients. A conservative interpretation in light of the quantitative bias analysis is that there is no evidence that NSAID use is associated with risk of increased severity or the other measured outcomes. Our results confirm and extend analogous findings in previous observational studies using a large cohort of patients drawn from 38 centers in a nationally representative multicenter database.

Motivated by the observation that network-based methods for the automatic prediction of protein functions can greatly benefit from exploiting both the similarity between proteins and the similarity between functional classes (as encoded, e.g., in the Gene Ontology), in this paper we propose a novel approach to the problem, based on the notion of a “graph transduction game.” We envisage a (non-cooperative) game, played over a graph, where the players (graph vertices) represent proteins, the functional classes correspond to the (pure) strategies, and protein- and function-level similarities are combined into a suitable payoff function. Within this formulation, Nash equilibria turn out to provide consistent functional labelings of proteins, and we use classical replicator dynamics from evolutionary game theory to find them. To test the effectiveness of our approach we conducted experiments on five different organisms and three ontologies, and the results obtained show that our method compares favorably with state-of-the-art algorithms.

Background Cis-regulatory regions (CRRs) are non-coding regions of the DNA that fine control the spatio-temporal pattern of transcription; they are involved in a wide range of pivotal processes such as the development of specific cell-lines/tissues and the dynamic cell response to physiological stimuli. Recent studies showed that genetic variants occurring in CRRs are strongly correlated with pathogenicity or deleteriousness. Considering the central role of CRRs in the regulation of physiological and pathological conditions, the correct identification of CRRs and of their tissue-specific activity status through Machine Learning methods plays a major role in dissecting the impact of genetic variants on human diseases. Unfortunately, the problem is still open, though some promising results have been already reported by (deep) machine-learning based methods that predict active promoters and enhancers in specific tissues or cell lines by encoding epigenetic or spectral features directly extracted from DNA sequences. Results We present the experiments we performed to compare two Deep Neural Networks, a Feed-Forward Neural Network model working on epigenomic features, and a Convolutional Neural Network model working only on genomic sequence, targeted to the identification of enhancer- and promoter-activity in specific cell lines. While performing experiments to understand how the experimental setup influences the prediction performance of the methods, we particularly focused on (1) automatic model selection performed by Bayesian optimization and (2) exploring different data rebalancing setups for reducing negative unbalancing effects. Conclusions Results show that (1) automatic model selection by Bayesian optimization improves the quality of the learner; (2) data rebalancing considerably impacts the prediction performance of the models; test set rebalancing may provide over-optimistic results, and should therefore be cautiously applied; (3) despite working on sequence data, convolutional models obtain performance close to those of feed forward models working on epigenomic information, which suggests that also sequence data carries informative content for CRR-activity prediction. We therefore suggest combining both models/data types in future works.

Inhibiting protein kinases (PKs) that cause cancers has been an important topic in cancer therapy for years. So far, almost 8% of >530 PKs have been targeted by FDA-approved medications, and around 150 protein kinase inhibitors (PKIs) have been tested in clinical trials. We present an approach based on natural language processing and machine learning to investigate the relations between PKs and cancers, predicting PKs whose inhibition would be efficacious to treat a certain cancer. Our approach represents PKs and cancers as semantically meaningful 100-dimensional vectors based on word and concept neighborhoods in PubMed abstracts. We use information about phase I-IV trials in ClinicalTrials.gov to construct a training set for random forest classification. Our results with historical data show that associations between PKs and specific cancers can be predicted years in advance with good accuracy. Our tool can be used to predict the relevance of inhibiting PKs for specific cancers and to support the design of well-focused clinical trials to discover novel PKIs for cancer therapy.

Inhibiting protein kinases (PKs) that cause cancers has been an important topic in cancer therapy for years. So far, almost 8% of more than 530 PKs have been targeted by FDA-approved medications and around 150 protein kinase inhibitors (PKIs) have been tested in clinical trials. We present an approach based on natural language processing and machine learning to the relations between PKs and cancers, predicting PKs whose inhibition would be efficacious to treat a certain cancer. Our approach represents PKs and cancers as semantically meaningful 100-dimensional vectors based on co-occurrence patterns in PubMed abstracts. We use information about phase I-IV trials inClinicalTrials.govto construct a training set for random forest classification. In historical data, associations between PKs and specific cancers could be predicted years in advance with good accuracy. Our model may be a tool to predict the relevance of inhibiting PKs with specific cancers.

We provide here a first-hand description of the coseismic surface effects caused by the Mw 6.4 Petrinja earthquake that hit central Croatia on 29 December 2020. This was one of the strongest seismic events that occurred in Croatia in the last two centuries. Field surveys in the epicentral area allowed us to observe and map primary coseismic effects, including geometry and kinematics of surface faulting, as well as secondary effects, such as liquefaction, sinkholes and landslides. The resulting dataset consists of homogeneous georeferenced records identifying 222 observation points, each of which contains a minimum of 5 to a maximum of 14 numeric and string fields of relevant information. The earthquake caused surface faulting defining a typical ‘conjugate’ fault pattern characterized by Y and X shears, tension cracks (T fractures), and compression structures (P shears) within a ca. 10 km wide (across strike), NW–SE striking right-lateral strike-slip shear zone (i.e., the Petrinja Fault Zone, PFZ). We believe that the results of the field survey provide fundamental information to improve the interpretation of seismological, GPS and InSAR data of this earthquake. Moreover, the data related to the surface faulting may impact future studies focused on earthquake processes in active strike-slip settings, integrating the estimates of slip amount and distribution in assessing the hazard associated with capable transcurrent faults.

Motivation Automated protein function prediction is a complex multi-class, multi-label, structured classification problem in which protein functions are organized in a controlled vocabulary, according to the Gene Ontology (GO). ‘Hierarchy-unaware’ classifiers, also known as ‘flat’ methods, predict GO terms without exploiting the inherent structure of the ontology, potentially violating the True-Path-Rule (TPR) that governs the GO, while ‘hierarchy-aware’ approaches, even if they obey the TPR, do not always show clear improvements with respect to flat methods, or do not scale well when applied to the full GO. Results To overcome these limitations, we propose Hierarchical Ensemble Methods for Directed Acyclic Graphs (HEMDAG), a family of highly modular hierarchical ensembles of classifiers, able to build upon any flat method and to provide ‘TPR-safe’ predictions, by leveraging a combination of isotonic regression and TPR learning strategies. Extensive experiments on synthetic and real data across several organisms firstly show that HEMDAG can be used as a general tool to improve the predictions of flat classifiers, and secondly that HEMDAG is competitive versus state-of-the-art hierarchy-aware learning methods proposed in the last CAFA international challenges. Availability and implementation Fully tested R code freely available at https://anaconda.org/bioconda/r-hemdag. Tutorial and documentation at https://hemdag.readthedocs.io. Supplementary information Supplementary data are available at Bioinformatics online.

We provide here a first-hand description of the coseismic surface effects caused by the Mw 6.4 Petrinja earthquake that hit central Croatia on 29 December 2020. This was one of the strongest seismic events that occurred in Croatia in the last two centuries. Field surveys in the epicentral area allowed us to observe and map primary coseismic effects, including geometry and kinematics of surface faulting, as well as secondary effects, such as liquefaction, sinkholes and landslides. The resulting dataset consists of homogeneous georeferenced records identifying 222 observation points, each of which contains a minimum of 5 to a maximum of 14 numeric and string fields of relevant information. The earthquake caused surface faulting defining a typical ‘conjugate’ fault pattern characterized by Y and X shears, tension cracks (T fractures), and compression structures (P shears) within a ca. 10 km wide, right-lateral strike-slip fault zone (i.e. the Petrinja Fault Zone, PFZ). We believe that the results of the field survey provide fundamental information to improve the interpretation of seismological, GPS and InSAR data of this earthquake. Moreover, the data related to the surface faulting may impact future studies focused on earthquake processes in active strike-slip settings, integrating the estimates of slip amount and distribution in assessing the hazard associated with capable transcurrent faults.

A critical challenge in genetic diagnostics is the computational assessment of candidate splice variants, specifically the interpretation of nucleotide changes located outside of the highly conserved dinucleotide sequences at the 5′ and 3′ ends of introns. To address this gap, we developed the Super Quick Informationcontent Random-forest Learning of Splice variants (SQUIRLS) algorithm. SQUIRLS generates a small set of interpretable features for machine learning by calculating the information-content (IC) of wildtype and variant sequences of canonical and cryptic splice sites, assessing changes in candidate splicing regulatory sequences, and incorporating characteristics of the sequence such as exon length, disruptions of the AG exclusion zone, and conservation. We curated a comprehensive collection of disease-associated splicealtering variants at positions outside of the highly conserved AG/GT dinucleotides at the termini of introns. SQUIRLS trains two random-forest classifiers for the donor and for the acceptor and combines their outputs by logistic regression to yield a final score. We show that SQUIRLS transcends previous state of the art accuracy in classifying splice variants as assessed by rank analysis in simulated exomes and is significantly faster than competing methods. SQUIRLS provides tabular output files for incorporation into diagnostic pipelines for exome and genome analysis, as well as visualizations that contextualize predicted effects of variants on splicing to make it easier to interpret splice variants in diagnostic settings

Spreadsheets are often used as a simple way for representing tabular data. However, since they do not impose any restriction on their table structures and contents, their automatic processing and the integration with other information sources are particularly hard problems to solve. Many table understanding approaches have been proposed for extracting data from tables and transforming them in meaningful information. However, they require some regularities on the table contents.

The visual exploration and analysis of biomolecular networks is of paramount importance for identifying hidden and complex interaction patterns among proteins. Although many tools have been proposed for this task, they are mainly focused on the query and visualization of a single protein with its neighborhood. The global exploration of the entire network and the interpretation of its underlying structure still remains difficult, mainly due to the excessively large size of the biomolecular networks. In this paper we propose a novel multi-resolution representation and exploration approach that exploits hierarchical community detection algorithms for the identification of communities occurring in biomolecular networks. The proposed graphical rendering combines two types of nodes (protein and communities) and three types of edges (protein-protein, community-community, protein-community), and displays communities at different resolutions, allowing the user to interactively zoom in and out from different levels of the hierarchy. Links among communities are shown in terms of relationships and functional correlations among the biomolecules they contain. This form of navigation can be also combined by the user with a vertex centric visualization for identifying the communities holding a target biomolecule. Since communities gather limited-size groups of correlated proteins, the visualization and exploration of complex and large networks becomes feasible on off-the-shelf computer machines. The proposed graphical exploration strategies have been implemented and integrated in UNIPred-Web, a web application that we recently introduced for combining the UNIPred algorithm, able to address both integration and protein function prediction in an imbalance-aware fashion, with an easy to use vertex-centric exploration of the integrated network. The tool has been deeply amended from different standpoints, including the prediction core algorithm. Several tests on networks of different size and connectivity have been conducted to show off the vast potential of our methodology; moreover, enrichment analyses have been performed to assess the biological meaningfulness of detected communities. Finally, a CoV-human network has been embedded in the system, and a corresponding case study presented, including the visualization and the prediction of human host proteins that potentially interact with SARS-CoV2 proteins.

Several problems in computational biology and medicine are modelled as learning problems in graphs, where nodes represent the biological entities to be studied, e.g. proteins, and connections dierent kinds of relationships among them, e.g. protein-protein interactions. In this context, classes are usually characterized by a high imbalance, i.e. positive examples for a class are much less than those negative. Although several works studied this problem, no graphbased software designed to explicitly take into account the label imbalance in biological networks is available. We propose COSNet, an R package to serve this purpose. COSNet deals with the label imbalance problem by implementing a novel parametric model of Hopeld Network (HN), whose output levels and activation thresholds of neurons are parameters to be automatically learnt. Due to the quasi linear time complexity, COSNet nicely scales when the number of instances is large, and application examples to challenging problems in biomedicine show the eciency and the accuracy of the proposed library.

The annotation and characterization of tissue-specific cis-regulatory elements (CREs) in non-coding DNA represents an open challenge in computational genomics. Several prior works show that machine learning methods, using epigenetic or spectral features directly extracted from DNA sequences, can predict active promoters and enhancers in specific tissues or cell lines. In particular, very recently deep-learning techniques obtained state-of-the-art results in this challenging computational task. In this study, we provide additional evidence that Feed Forward Neural Networks (FFNN) trained on epigenetic data and one-dimensional convolutional neural networks (CNN) trained on DNA sequence data can successfully predict active regulatory regions in different cell lines. We show that model selection by means of Bayesian optimization applied to both FFNN and CNN models can significantly improve deep neural network performance, by automatically finding models that best fit the data. Further, we show that techniques applied to balance active and non-active regulatory regions in the human genome in training and test data may lead to over-optimistic or poor predictions. We recommend to use actual imbalanced data that was not used to train the models for evaluating their generalization performance.

Multitask algorithms typically use task similarity information as a bias to speed up and improve the performance of learning processes. Tasks are learned jointly, sharing information across them, in order to construct models more accurate than those learned separately over single tasks. In this contribution, we present the first multitask model, to our knowledge, based on Hopfield Networks (HNs), named HoMTask. We show that by appropriately building a unique HN embedding all tasks, a more robust and effective classification model can be learned. HoMTask is a transductive semi-supervised parametric HN, that minimizes an energy function extended to all nodes and to all tasks under study. We provide theoretical evidence that the optimal parameters automatically estimated by HoMTask make coherent the model itself with the prior knowledge (connection weights and node labels). The convergence properties of HNs are preserved, and the fixed point reached by the network dynamics gives rise to the prediction of unlabeled nodes. The proposed model improves the classification abilities of singletask HNs on a preliminary benchmark comparison, and achieves competitive performance with state-of-the-art semi-supervised graph-based algorithms.

Between January and October of 2020, the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus has infected more than 34 million persons in a worldwide pandemic leading to over one million deaths worldwide (data from the Johns Hopkins University). Since the virus begun to spread, emergency departments were busy with COVID-19 patients for whom a quick decision regarding in- or outpatient care was required. The virus can cause characteristic abnormalities in chest radiographs (CXR), but, due to the low sensitivity of CXR, additional variables and criteria are needed to accurately predict risk. Here, we describe a computerized system primarily aimed at extracting the most relevant radiological, clinical, and laboratory variables for improving patient risk prediction, and secondarily at presenting an explainable machine learning system, which may provide simple decision criteria to be used by clinicians as a support for assessing patient risk. To achieve robust and reliable variable selection, Boruta and Random Forest (RF) are combined in a 10-fold cross-validation scheme to produce a variable importance estimate not biased by the presence of surrogates. The most important variables are then selected to train a RF classifier, whose rules may be extracted, simplified, and pruned to finally build an associative tree, particularly appealing for its simplicity. Results show that the radiological score automatically computed through a neural network is highly correlated with the score computed by radiologists, and that laboratory variables, together with the number of comorbidities, aid risk prediction. The prediction performance of our approach was compared to that that of generalized linear models and shown to be effective and robust. The proposed machine learning-based computational system can be easily deployed and used in emergency departments for rapid and accurate risk prediction in COVID-19 patients.

The Automated Functional Prediction (AFP) of proteins became a challenging problem in bioinformatics and biomedicine aiming at handling and interpreting the extremely large-sized proteomes of several eukaryotic organisms. A central issue in AFP is the absence in public repositories for protein functions, e.g. the Gene Ontology (GO), of well defined sets of negative examples to learn accurate classifiers for AFP. In this paper we investigate the Query by Committee paradigm of active learning to select the negatives most informative for the classifier and the protein function to be inferred. We validated our approach in predicting the Gene Ontology function for the S.cerevisiae proteins.

Many interactions among bio-molecular entities, e.g. genes, proteins, metabolites, can be easily represented by means of property graphs, i.e. graphs that are annotated both on the vertices (e.g. entity identifier, Gene Ontology or Human Phenotype Ontology terms) and on the edges (the strength of the relationship, the evidence of the source from which the weight has been taken, etc.). These graphs contain a relevant information that can be exploited for conducting different kinds of analysis, such as automatic function prediction, disease gene prioritization, drug repositioning. However, the number and size of the networks are becoming quite large and there is the need of tools that allow the biologists to manage the networks, graphically explore their structures, and organize the visualization and analysis of the graph according to different perspectives. In this paper we introduce the web service that we have developed for the visual analysis of biomolecular networks. Specifically we will show the different functionalities for exploring big networks (that do not fit in the current canvas) starting from a specific vertex, for changing the view perspective of the network, and for navigating the network and thus identifying new relationships. The proposed system extends the functionalities of off-the-shelf graphical visualization tools (e.g. GraphViz and GeneMania) by limiting the production of big cloud of points and allowing further customized visualizations of the network and introducing their vertex-centric exploration.

The attribute-based access control (ABAC) model has been gaining popularity in recent years because of its advantages in granularity, flexibility, and usability. Few approaches based on association rules mining have been proposed for the automatic generation of ABAC policies from access logs. Their aim is the identification of policies that do not overfit over training data, are not too general and thus does to disclose sensitive resources to everyone, and are interpretable by humans. The large ABAC privilege space along with the sparsity and unbalance distribution of the available logs make the solution of this task particularly complex and current approaches have different limitations. In this paper we compare different symbolic and nonsymbolic machine learning (ML) techniques for inferring ABAC policies and discuss their pros and cons. Based on experimental results on a toy dataset and on a real dataset, we argue that which is the best technique depends on the characteristics of the considered data. When the data are highly separable according to PCA and t-SNE decomposition, the quality of the obtained ABAC policies is higher and also policies are easily interpretable. By contrast, when this property does not hold, the quality of the obtained policies is low; in this case, non-symbolic ML techniques show better results than the symbolic ones.

One of the main issues in detecting the genes involved in the etiology of genetic human diseases is the integration of different types of available functional relationships between genes. Numerous approaches exploited the complementary evidence coded in heterogeneous sources of data to prioritize disease-genes, such as functional profiles or expression quantitative trait loci, but none of them to our knowledge posed the scarcity of known disease-genes as a feature of their integration methodology. Nevertheless, in contexts where data are unbalanced, that is, where one class is largely under-represented, imbalance-unaware approaches may suffer a strong decrease in performance. We claim that imbalance-aware integration is a key requirement for boosting performance of gene prioritization (GP) methods. To support our claim, we propose an imbalance-aware integration algorithm for the GP problem, and we compare it on benchmark data with other state-of-the-art integration methodologies.

The Human Phenotype Ontology (HPO) provides a standard categorization of the phenotypic abnormalities encountered in human diseases and of the semantic relationship between them. Quite surprisingly the problem of the automated prediction of the association between genes and abnormal human phenotypes has been widely overlooked, even if this issue represents an important step toward the characterization of gene-disease associations, especially when no or very limited knowledge is available about the genetic etiology of the disease under study. We present a novel ensemble method able to capture the hierarchical relationships between HPO terms, and able to improve existing hierarchical ensemble algorithms by explicitly considering the predictions of the descendant terms of the ontology. In this way the algorithm exploits the information embedded in the most specific ontology terms that closely characterize the phenotypic information associated with each human gene. Genome-wide results obtained by integrating multiple sources of information show the effectiveness of the proposed approach.

In several real-world node label prediction problems on graphs, in fields ranging from computational biology to World Wide Web analysis, nodes can be partitioned into categories different from the classes to be predicted, on the basis of their characteristics or their common properties. Such partitions may provide further information about node classification that classical machine learning algorithms do not take into account. We introduce a novel family of parametric Hopfield networks (m-category Hopfield networks) and a novel algorithm (Hopfield multi-category--HoMCat), designed to appropriately exploit the presence of property-based partitions of nodes into multiple categories. Moreover, the proposed model adopts a cost-sensitive learning strategy to prevent the remarkable decay in performance usually observed when instance labels are unbalanced, that is, when one class of labels is highly underrepresented than the other one. We validate the proposed model on both synthetic and real-world data, in the context of multi-species function prediction, where the classes to be predicted are the Gene Ontology terms and the categories the different species in the multi-species protein network. We carried out an intensive experimental validation, which on the one hand compares HoMCat with several state-of-the-art graph-based algorithms, and on the other hand reveals that exploiting meaningful prior partitions of input data can substantially improve classification performances.

Summary: RANKS is a flexible software package that can be easily applied to any bioinformatics task formalizable as ranking of nodes with respect to a property given as a label, such as automated protein function prediction, gene disease prioritization and drug repositioning. To this end RANKS provides an efficient and easy-to-use implementation of kernelized score functions, a semi-supervised algorithmic scheme embedding both local and global learning strategies for the analysis of biomolecular networks. To facilitate comparative assessment, baseline network-based methods, e.g. label propagation and random walk algorithms, have also been implemented. Availability and Implementation: The package is available from CRAN: https://cran.r-project.org/. The package is written in R, except for the most computationally intensive functionalities which are implemented in C. Contact: valentini@di.unimi.it Supplementary information: Supplementary data are available at Bioinformatics online.

A semiquantitative analysis by means of midinfrared FT-IR spectroscopy was tuned for the simultaneous determination of cellulose, hemicellulose, and lignin in industrial crops such as giant reed (Arundo donaxL.) and switchgrass (Panicum virgatumL.). Ternary mixtures of pure cellulose, hemicellulose, and lignin were prepared and a direct correlation area/concentration was achieved for cellulose and lignin, whereas indirect correlations were found for hemicellulose quantification. Good correspondences between the values derived from our model and those reported in the literature or obtained according to the official Van Soest method were ascertained. Average contents of 40–45% of cellulose, 20–25% of hemicellulose, and 20–25% of lignin were obtained for different samples of giant reed species. In the case of switchgrass, a content of 36% of cellulose, 28% of hemicellulose, and 26% of lignin was achieved. This analysis was also carried out on giant reed and switchgrass residues after a mild hydrolysis step carried out with dilute hydrochloric acid for the production of furfural with good yield. Reasonable compositional data were obtained, thus allowing an indirect monitoring which helps the optimization of the hydrothermal pretreatment for furfural production from hemicellulose fractions.

The regulatory code that determines whether and how a given genetic variant affects the function of a regulatory element remains poorly understood for most classes of regulatory variation. Indeed the large majority of bioinformatics tools have been developed to predict the pathogenicity of genetic variants in coding sequences or conserved splice sites. Computational algorithms for the prediction of non-coding deleterious variants associated with rare genetic diseases are faced with special challenges owing to the rarity of confirmed pathogenic mutations. Indeed in this context classical machine learning methods are biased toward neutral variants that constitute the large majority of genetic variation, and are not able to detect the potential deleterious variants that constitute only a tiny minority of all known genetic variation. We recently proposed hyperSMURF, hyper-ensemble of SMOTE Undersampled Random Forests, an ensemble approach explicitly designed to deal with the huge imbalance between deleterious and neutral variants, and able to significantly outperform state-of-the-art methods for the prediction of non-coding variants associated with Mendelian diseases. Despite its successful application to the detection of deleterious single nucleotide variants (SNV) as well as to small insertions or deletions (indels), hyperSMURF is a method that depends on several learning parameters, that strongly influence its overall performances. In this work we experimentally show that by tuning hyperSMURF parameters we can significantly boost the performance of the method, thus predicting with significantly better precision and recall rare SNVs associated with Mendelian diseases.

Protein function prediction is a complex multiclass multilabel classification problem, characterized by multiple issues such as the incompleteness of the available annotations, the integration of multiple sources of high dimensional biomolecular data, the unbalance of several functional classes, and the difficulty of univocally determining negative examples. Moreover, the hierarchical relationships between functional classes that characterize both the Gene Ontology and FunCat taxonomies motivate the development of hierarchy-aware prediction methods that showed significantly better performances than hierarchical-unaware “flat” prediction methods. In this paper, we provide a comprehensive review of hierarchical methods for protein function prediction based on ensembles of learning machines. According to this general approach, a separate learning machine is trained to learn a specific functional term and then the resulting predictions are assembled in a “consensus” ensemble decision, taking into account the hierarchical relationships between classes. The main hierarchical ensemble methods proposed in the literature are discussed in the context of existing computational methods for protein function prediction, highlighting their characteristics, advantages, and limitations. Open problems of this exciting research area of computational biology are finally considered, outlining novel perspectives for future research.

Drug repositioning is a challenging computational problem involving the integration of heterogeneous sources of biomolecular data and the design of label ranking algorithms able to exploit the overall topology of the underlying pharmacological network. In this context, we propose a novel semisupervised drug ranking problem: prioritizing drugs in integrated biochemical networks according to specific DrugBank therapeutic categories. Algorithms for drug repositioning usually perform the inference step into an inhomogeneous similarity space induced by the relationships existing between drugs and a second type of entity (e.g., disease, target, ligand set), thus making unfeasible a drug ranking within a homogeneous pharmacological space. To deal with this problem, we designed a general framework based on bipartite network projections by which homogeneous pharmacological networks can be constructed and integrated from heterogeneous and complementary sources of chemical, biomolecular and clinical information. Moreover, we present a novel algorithmic scheme based on kernelized score functions that adopts both local and global learning strategies to effectively rank drugs in the integrated pharmacological space using different network combination methods. Detailed experiments with more than 80 DrugBank therapeutic categories involving about 1,300 FDA-approved drugs show the effectiveness of the proposed approach.

Significant advances in high-throughput sequencing technologies raised exponentially the rate of acquisition of novel biological knowledge in the last decade, thus resulting in consistent difficulties in the analysis of vast amount of biological data. This adverse scenario is exacerbated by serious scalability limitations affecting state-of-the art within-network learning methods and by the limited availability of primary memory in off-the-shelf desktop computers. In this contribution we present the application of a novel graph kernel, transductive and secondary memory-based network learning algorithm able to effectively tackle the aforementioned limitations. The proposed algorithm is then evaluated on a large (more than 200,000 vertices) biological network using ordinary off-the-shelf computers. To our knowledge this is the first time a graph kernel learning method is applied to a so large biological network.

The visualization and analysis of big bio-molecular networks is a key feature for the investigation and prediction of protein functions in a multi-species context. In this paper we present the design of a system that integrates data management, machine learning and visualization facilities to make effective the visual analysis of big networks by means of web-based interfaces.

The removal of polychlorinated biphenyls (PCBs) both from siloxane transformer oil and hydrocarbon engine oil was investigated through the application of microwave (MW) irradiation and a reaction system based on polyethyleneglycol (PEG) and potassium hydroxide. The influence of the main reaction parameters (MW irradiation time, molecular weight of PEG, amount of added reactants and temperature) on the dechlorination behavior was studied. Promising performances were reached, allowing about 50% of dechlorination under the best experimental conditions, together time and energy saving compared to conventional heating systems. Moreover, an interesting dechlorination degree (up to 32%) was achieved for siloxane transformer oil when MW irradiation was employed as the unique driving force. To the best of our knowledge, this is the first time in which MW irradiation is tested as the single driving force for the dechlorination of these two types of PCB-contaminated oils.

Gene function prediction is a complex multilabel classification problem with several distinctive features: the hierarchical relationships between functional classes, the presence of multiple sources of biomolecular data, the unbalance between positive and negative examples for each class, the complexity of the whole-ontology and genome-wide dimensions. Unlike previous works, which mostly looked at each one of these issues in isolation, we explore the interaction and potential synergy of hierarchical multilabel methods, data fusion methods, and cost-sensitive approaches on whole-ontology and genome-wide gene function prediction. Besides classical top-down hierarchical multilabel ensemble methods, in our experiments we consider two recently proposed multilabel methods: one based on the approximation of the Bayesian optimal classifier with respect to the hierarchical loss, and one based on a heuristic approach inspired by the true path rule for the biological functional ontologies. Our experiments show that key factors for the success of hierarchical ensemble methods are the integration and synergy among multilabel hierarchical, data fusion, and cost-sensitive approaches, as well as the strategy of selecting negative examples.

Summary The availability of various high-throughput experimental and computational methods developed in the last decade allowed molecular biologists to investigate the functions of genes at system level opening unprecedented research opportunities. Despite the automated prediction of genes functions could be included in the most difficult problems in bioinformatics, several recently published works showed that consistent improvements in prediction performances can be obtained by integrating heterogeneous data sources. Nevertheless, very few works have been dedicated to the investigation of the impact of noisy data on the prediction performances achievable by using data integration approaches.In this contribution we investigated the tolerance of multiple classifier systems (MCS) to noisy data in gene function prediction experiments based on data integration methods. The experimental results show that performances of MCS do not undergo a significant decay when noisy data sets are added. In addition, we show that in this task MCS are competitive with kernel fusion, one of the most widely applied technique for data integration in gene function prediction problems.

High-throughput biotechnologies are playing an increasingly important role in biomolecular research. Their ability to provide genome wide views of molecular mechanisms occurring in living cells could play a crucial role in the elucidation of biomolecular processes at system level but dataset produced using these techniques are often high-dimensional and very noisy making their analysis challenging because the need to extract relevant information froma sea of noise. Gene function prediction is a central problem in modern bioinformatics and recent works pointed out that gene function prediction performances can be improved by integrating heterogeneous biomolecular datasources. In this contribution we compared performances achievable in gene function prediction by early and late data fusion methods. Given that, among the available late fusion methods, ensemble systems have not been, at today, extensively investigated, all the late fusion experiments were performed using multiple classifier systems. Experimental results show that late fusion of heterogeneous datasets realized by mean of ensemble systems outperformed both early fusion approaches and base learners trained on single types of biomolecular data.

Several solutions have been proposed to exploit the availability of heterogeneous sources of biomolecular data for gene function prediction, but few attention has been dedicated to the evaluation of the potential improvement in functional classification results that could be achieved through data fusion realized by means of ensemble-based techniques. In this contribution we test the performance of several ensembles of support vector machine (SVM) classifiers, in which each component learner has been trained on different types of bio-molecular data, and then combined to obtain a consensus prediction using different aggregation techniques. Experimental results using data obtained with different high-throughput biotechnologies show that simple ensemble methods outperform both learning machines trained on single homogeneous types of bio-molecular data, and vector space integration methods.

The analysis of non-coding DNA regulatory regions is one of the most challenging open problems in computational biology. In this paper we investigate whether we can predict functional information about genes by using information extracted from their sequences together with expression data. We formalize this problem as a classification problem, and we apply Support Vector Machines (SVMs) with non-linear kernels to predict classes of co-expressed genes obtained from clustering procedures. SVMs are trained using information about selected motifs extracted from DNA regulatory regions through combinatorial and statistical methods. In our experiments, we show that functional classes of genes can be predicted from biological sequence data in Saccharomices cerevisiae, achieving results competitive with those recently presented in the literature.

Objective:: Clustering algorithms may be applied to the analysis of DNA microarray data to identify novel subgroups that may lead to new taxonomies of diseases defined at bio-molecular level. A major problem related to the identification of biologically meaningful clusters is the assessment of their reliability, since clustering algorithms may find clusters even if no structure is present. Methodology:: Recently, methods based on random ''perturbations'' of the data, such as bootstrapping, noise injections techniques and random subspace methods have been applied to the problem of cluster validity estimation. In this framework, we propose stability measures that exploits the high dimensionality of DNA microarray data and the redundancy of information stored in microarray chips. To this end we randomly project the original gene expression data into lower dimensional subspaces, approximately preserving the distance between the examples according to the Johnson-Lindenstrauss (JL) theory. The stability of the clusters discovered in the original high dimensional space is estimated by comparing them with the clusters discovered in randomly projected lower dimensional subspaces. The proposed cluster-stability measures may be applied to validate and to quantitatively assess the reliability of the clusters obtained by a large class of clustering algorithms. Results and conclusion:: We tested the effectiveness of our approach with high dimensional synthetic data, whose distribution is a priori known, showing that the stability measures based on randomized maps correctly predict the number of clusters and the reliability of each individual cluster. Then we showed how to apply the proposed measures to the analysis of DNA microarray data, whose underlying distribution is unknown. We evaluated the validity of clusters discovered by hierarchical clustering algorithms in diffuse large B-cell lymphoma (DLBCL) and malignant melanoma patients, showing that the proposed reliability measures can support bio-medical researchers in the identification of stable clusters of patients and in the discovery of new subtypes of diseases characterized at bio-molecular level.

The problem of assessing the reliability of clusters patients identified by clustering algorithms is crucial to estimate the significance of subclasses of diseases detectable at bio-molecular level, and more in general to support bio-medical discovery of patterns in gene expression data. In this paper we present an experimental analysis of the reliability of clusters discovered in lung tumor patients using DNA microarray data. In particular we investigate if subclasses of lung adenocarcinoma can be detected with high reliability at bio-molecular level. To this end we apply cluster validity measures based on random projections recently proposed by Bertoni and coworkers. The results show that at least two subclasses of lung adenocarcinoma can be detected with relatively high reliability, confirming and extending previous findings reported in the literature.

Support vector machines (SVMs), and other supervised learning techniques have been experimented for the bio-molecular diagnosis of malignancies, using also feature selection methods. The classification task is particularly difficult because of the high dimensionality and low cardinality of gene expression data. In this paper we investigate a different approach based on random subspace ensembles of SVMs: a set of base learners is trained and aggregated using subsets of features randomly drawn from the available DNA microarray data. Experimental results on the colon adenocarcinoma diagnosis and medulloblastoma clinical outcome prediction show the effectiveness of the proposed approach.

One of the main factors a3ecting the e3ectiveness of error correcting output coding (ECOC) methods for classi4cation is the dependence among the errors of the computed codeword bits. We present an extensive experimental work for evaluating the dependence among output errors of the decomposition unit in ECOC learning machines. In particular, we apply measures based on mutual information to compare the dependence of ECOC multi-layer perceptron (ECOC MLP), made up by a single multi-input multi-output MLP, and ECOC ensembles made up by a set of independent and parallel dichotomizers (ECOC PND). Moreover, the experimentation analyzes the relationship between the architecture, the dependence among output errors and the performances of ECOC learning machines. The results show that the dependence among computed codeword bits is signi4cantly smaller for ECOC PND, pointing out that ensembles of independent parallel dichotomizers are better suited for implementing ECOC classi4cation methods. The experimental results suggest new architectures of ECOC learning machines to improve the independence among output errors and the diversity between base learners. c

Structured taxonomies characterize several real world problems, ranging from text categorization, to video annotation and protein function prediction. In this context “flat” learning methods may introduce inconsistent predictions, while structured output-aware learning methods can improve the accuracy of the predictions by exploiting the hierarchical relationships between classes. We propose a novel hierarchical ensemble method able to provide theoretically guaranteed consistent predictions for any Directed Acyclic Graph (DAG)-structured taxonomy, and consequently also for any taxonomy structured according to a tree. Results with a complex real-world DAG-structured taxonomy involving about one thousand classes and twenty thousand of examples show that the proposed hierarchical ensemble approach significantly improves flat methods, especially in terms of precision/recall curves.