Accurate characterization of breast lesions remains a challenge in oncological imaging. Contrast-Enhanced Spectral Mammography (CESM) has recently emerged as a cost-effective alternative to magnetic resonance imaging (MRI), providing both low-energy (LE) and subtracted (SUB) images. While histopathology remains the diagnostic reference, growing evidence suggests that morphological and functional information extracted from CESM can support the discrimination between malignant and benign lesions. In this paper, we present a reproducible multiview learning pipeline for CESM, designed to evaluate three fusion strategies -early, late, and hybrid- across both LE and SUB images. Experiments conducted on the public CDD-CESM dataset of 326 patients show that single-view models on SUB images achieve the best performance (Acc$=84.5 \%$, AUC$=88.7 \%$), whereas multiview fusion does not consistently improve over the strongest single-view baseline. These results confirm the high discriminative power of SUB images and suggest that naive fusion may add limited value. Our contribution is to provide the first transparent benchmark of CESM fusion strategies on a public dataset, enabling like-for-like comparisons across future studies. From a clinical perspective, this reproducible framework may inform the design of decision-support tools, with the potential to reduce unnecessary biopsies and assist radiologists in challenging diagnostic settings.

Network alignment (NA) is a computational methodology employed to compare biological networks across different species or conditions. By identifying conserved structures, functions, and interactions, NA provides invaluable insights into shared biological processes, evolutionary relationships, and system-level behaviors. This manuscript presents a comprehensive overview of NA methodologies, including the importance of preprocessing network data, selecting suitable input formats, and understanding diverse network types such as attributed, temporal, and multilayer networks. Additionally, it explores key challenges such as seed nodes selection, algorithm configuration, and cross-species alignment, emphasizing the necessity of integrating functional annotations, sequence similarity, and network topology for biologically meaningful results. Various NA strategies, including Local and Global Network Alignment, are discussed alongside their respective advantages and limitations. Practical recommendations for effectively documenting and visualizing NA experiments are also provided, ensuring reproducibility and clarity in research. By leveraging diverse alignment tools and adopting best practices, researchers can unlock the potential of NA to advance our understanding of complex biological systems.

Breast cancer (BC) remains a leading global malignancy, with taxane-based treatment (TBT), including paclitaxel and docetaxel, significantly enhancing outcomes across early-stage, locally advanced, and metastatic BC. Despite its efficacy, TBT can cause severe taxane-related peripheral neurotoxicity (TrPN), which limits dosing and affects patient quality of life. TrPN is cumulative, primarily sensory, and unpredictable, with genetic predisposition playing a key role in its development. Previous pharmacogenomic (PGx) study identified five single nucleotide polymorphisms (SNPs) in the NR1I3 and UGT2B7 genes, linked to protection against severe TrPN in BC patients. ROC analysis validated these findings in an independent BC dataset. Neuroprotective effects were in patients homozygous for allele variants 2 with ultrametabolizer phenotype, which accelerates taxane inactivation and reduces treatment efficacy, potentially worsening prognosis. NR1I3 and UGT2B7 genes could serve as predictive biomarkers for TrPN and taxane bioavailability. Further network and pathway enrichment analyses (PEA) provided insights into the molecular pathways involved in TrPN and BC progression. Network analysis, a powerful tool in computational biology, integrates genomic and proteomic data to uncover drug-response mechanisms. This approach advances personalized medicine by identifying key genes and biomarkers for predicting treatment response and adverse effects, offering new therapeutic opportunities.

In this paper, we aim to describe a novel deep learning model (a machine learning subclass) to classify fMRI. We used a publicly available dataset to train and test the model which involved patients affected by depression.Our model is based on a specific deep neural network, the Graph Attention Network (GAT) which has proven its strength in dealing with graph data representation. The novelty of our approach is that it is based on the extraction from the original fMRIs of graph representation then passed to the deep learning model.We performed this crucial phase by using a Matlab based toolbox, CONN, which helped in data preparation and graph representation extraction. We then used the extracted fMRI representations to feed, train, and finally test our deep learning model.While classification results were encouraging, achieving approximately 73% accuracy, another aspect that we investigated was focused on the comparison of three architectural solutions, focusing on power consumption. We used an Apple Silicon platform compared to a NVIDIA based laptop and an edge device of the NVIDIA Jetson family.

Networks and their properties have been used to study complex biological systems. Recently, network curvature measures have demonstrated the ability to capture relevant network properties. This study employs network curvature measures to analyse gene expression correlations in various human tissues for identifying unique topological features that differentiate these groups. Preliminary findings suggest that curvature measures offer novel insights that could enhance our understanding of the biological systems.

Multilayer networks (MLNs) have emerged as a critical tool in the field of medicine, particularly in neuroscience, owing to their capacity to model the complex interactions between brain regions across multiple dimensions. Unlike traditional single-layer network approaches, which typically focus on functional or structural connectivity within a single frequency band, MLN provides a richer, more comprehensive framework that captures the dynamic and multi-frequency nature of brain activity. In this work, we propose the development of a pipeline for the design and analysis of multilayer brain networks based on electroencephalogram (EEG) data. The primary object is to explore how MLNs can be utilized to analyze brain activity by capturing both intra- and inter-frequency interactions, that coordinate the different neural processes. The EEG data used in this study come from a cohort of 75 patients, including 25 healthy subjects, 25 with psychogenic non-epileptic seizures (PNES), and 25 with epilepsy. The results revealed significant differences in both the structure of the graphical network representation and the multilayer network analysis across the groups studied. These findings underscore the potential of multilayer networks (MLNs) to offer valuable insights into the distinct network patterns associated with various neurological conditions, providing a promising framework for advancing research into the complexity of brain network interactions.

Inflammatory Bowel Diseases (IBD), including Crohn’s Disease (CD) and Ulcerative Colitis (UC), are chronic conditions characterized by a complex network of comorbidities that significantly impact patients’ quality of life.In this study, we employed a Network Visualization approach to explore the comorbidities associated with IBD, using data extracted from the clinical records of patients at the Digestive Pathophysiology Unit of the University Hospital “Renato Dulbecco”.By constructing networks, we mapped and compared the comorbidities of Crohn’s Disease and Ulcerative Colitis, considering gender differences as well. The graphs produced offer a clear view clearly shows the major comorbidities and their distribution, highlighting significant differences between the two diseases and across genders.This visual approach facilitates a better understanding of the complexity of clinical interactions in IBD, suggesting new perspectives for personalized and multidisciplinary therapeutic approaches in managing these diseases.

BACKGROUND: The rewiring of molecular interactions in various conditions leads to distinct phenotypic outcomes. Differential network analysis (DINA) is dedicated to exploring these rewirings within gene and protein networks. Leveraging statistical learning and graph theory, DINA algorithms scrutinize alterations in interaction patterns derived from experimental data. RESULTS: Introducing a novel approach to differential network analysis, we incorporate differential gene expression based on sex and gender attributes. We hypothesize that gene expression can be accurately represented through non-Gaussian processes. Our methodology involves quantifying changes in non-parametric correlations among gene pairs and expression levels of individual genes. CONCLUSIONS: Applying our method to public expression datasets concerning diabetes mellitus and atherosclerosis in liver tissue, we identify gender-specific differential networks. Results underscore the biological relevance of our approach in uncovering meaningful molecular distinctions.

Healthcare data breaches are a growing problem that seriously threatens patient privacy, the reputation and trustworthiness of both public and private healthcare organizations. The aim of this paper is to elucidate the severity of healthcare data breaches, their potential impact on patient privacy and healthcare organizations, providing a predictive data breaches transformer conceptual architecture that can monitoring users actions that can result in possible system security violation and consequently in data breaches. At this regard, we introduce the description of a concept architecture for implementing a predictive data breaches transformer highlighting weakness and strengthens, and in the same time how the adoption of a predictive system can significantly limit the risk of the onset of possible data breaches by making the operator more aware in carrying out his activity in the processing of each type of data including personal health data.

Networks are present in different aspects of our life: communication networks, World Wide Web, Social Networks, and can be used to conveniently describe biological and clinical data, such as the interactions of proteins in an organism or the connections of neurons in the brain. Therefore, network science, focusing on the network representations of physical, biological and social phenomena and leading to predictive models of these phenomena, currently represents a vast field of application and research for many scientific and social disciplines. The mathematical background for the study and analysis of networks has its roots in the theory of graphs that allows studying real phenomena in a quantitative way. According to the formalism coming from graph theory, nodes of the graph represent entities, whereas edges represent the associations among them. Currently, in bioinformatics and systems biology, there is a growing interest in analyzing associations among biological molecules at a network level. Since the study of associations in a system-level scale has shown great potential, the use of networks has become the de facto standard for representing such associations, and its application fields span from molecular biology to brain connectome analysis [1]. Molecules of different types, e.g. genes, proteins, ribonucleic acids and metabolites, have fundamental roles in the mechanisms of the cellular processes. The study of their structure and interactions is crucial for different reasons, comprising the development of new drugs and the discovery of disease pathways. Thus, the modeling of the complete set of interactions and associations among biological molecules as a graph is convenient for a variety of reasons. Networks provide a simple and intuitive representation of heterogeneous and complex biological processes. Moreover, they facilitate modeling and understanding of complicated molecular mechanisms combining graph theory, machine learning and deep learning techniques. While proteomics and genomics data, represented as data streams or data tables, are mainly used to screen large populations in case–control studies (e.g. for early detection of diseases), interactomics data are represented as graphs and they add a new dimension of analysis, allowing, for instance, the graph-based comparison of organism’s properties. In general, complex biological systems represented as networks provide an integrated way to look into the dynamic behavior of the cellular system through the interactions of components. For instance, biological networks also referred to as Protein–Protein Interaction Networks, model biochemical interactions among proteins. Nodes represent the proteins from a given organism, and the edges represent the protein–protein interactions [2]. Also, gene regulatory network (GRN) is a collection of genes in a cell, which interact each other and with other substances in the cell, such as proteins or metabolites, thereby governing the rates at which genes in the network are transcribed into mRNA. Similarly, the graph-based modeling of the whole system of the brain elements and their relations, so-called brain connectome, is based on the representation of the regions of interest as nodes, and the representation of functional or anatomical connections as edges [3]. Furthermore, recent discoveries in biology have elucidated that the interplay of molecules of different types (e.g. genes, proteins and ribonucleic acids) is a constitutive block of mechanisms inside cells. Consequently, models describing the interplay should be able to consider the presence of multiple different agents and associations, i.e. multiple different types of nodes and edges, that yield to the so-called heterogeneous networks [4]. Networks and network analysis methods are a keystone in computational biology and bioinformatics and are increasingly used to study biological and clinical data in an integrated way. Network analysis consists of a collection of techniques with a shared methodological perspective, which allows to depict relations among entities and to analyze the structures that emerge from the recurrence of these relations. The basic assumption is that better explanations of different phenomena are yielded by the analysis of the relations among entities. Network analysis can be performed on networks built starting from omics data with the goal of extracting topological properties of the graph. These properties are then used to infer knowledge. For example, in interactomics, the identification of small subgraphs that are statistically overrepresented can be used to identify functionally relevant modules. Similarly, network analysis is used to highlight highly connected regions, assuming they can encode protein complexes. In addition, the systematic study of complex interactions among molecular components (i.e. DNA, RNA, microRNA, proteins and small molecules) is a new paradigm for discovering functional pathways on a global scale [5]. Finally, the network analysis can be conducted on clinical and biomedical data for investigating diseases. This Special Issue aims to collect relevant scientific contributions on fundamental network analysis methods and their applications in computational biology, bioinformatics and medicine. In particular, the Special Issue comprises contributions focusing both on networks modeling and analysis of omics data, as well as on modeling and analysis of clinical data. In Detection of pan-cancer surface protein biomarkers via a network-based approach on transcriptomics data, Daniele Mercatelli et al. [6] present a new network-based analysis protocol for transcriptomic data, in particular, focusing on the overall activity of curated surface proteins, with the final aim to identify those proteins driving major phenotype changes at a network level. The authors have demonstrated that their protocol is able to extract relevant knowledge within and across cancer data sets, by allowing to identify cancer-wide markers to design targeted therapies and biomarker-based diagnostic approaches. In Pathway integration and annotation: building a puzzle with non-matching pieces and no reference picture, Agapito et al. [7] present a review on the current methods for pathway consolidation. The authors start from the consideration that the absence of gold standards for pathway definition and representation as networks has led to the lack of overlap across databases and the lack of data integration across pathway databases. So the authors tackle the strengths and pitfalls of the current pathway consolidation methodologies and highlight directions for future improvements to this research area. In A generic parallel framework for inferring large-scale gene regulatory networks from expression profiles: application to Alzheimer’s disease network, Sebastian, et al. [8] present a framework that incorporates state-of-the-art methods as a black box, to infer GRN from expression profiles. The authors present a case study on the application of the framework to infer an Alzheimer’s disease-affected network from large expression profiles. On the other hand, the last article of the Special Issue focuses on networks modeling and analysis of clinical data. In Intersection of network medicine and machine learning towards investigating the key biomarkers and pathways underlying amyotrophic lateral sclerosis: a systematic review, Das et al. [9] present a review of the main network medicine approaches and implementations of network-based machine learning algorithms in amyotrophic lateral sclerosis, with the aim to identify critical pathways and biomarkers and therapeutic targets for personalized treatment. In summary, although investigating diseases through biological and biomedical data is continuously evolving, we hope this Special Issue will represent an authoritative and valuable resource for researchers. The editors are grateful to both the editor-in-chief and the publisher for having sustained this project, for their timely help and for having supported them in the day-to-day needs. A special thank is addressed to all the authors and reviewers, whose competence and effort allowed the realization of this Special Issue. Marianna Milano is an assistant professor and a senior research scientist in the field of omics data and biological networks analysis at the University ‘Magna Græcia’ of Catanzaro, Italy. Her research interests are focused on: the development of innovative algorithms for the analysis of clinical and omics data through the application of biological knowledge formalized in ontologies; the extraction of knowledge from biological and biomedical data; the use of formal knowledge representation tools in the field of computational biology; the development of algorithms for the analysis of biological and biomedical networks through the application of graph theory. She published one book and more than 60 papers in international journals and conference proceedings. She is a member of Bioinformatics Italian Society (BITS). Mario Cannataro is a full professor of computer engineering and the director of the Data Analytics research center at the University ‘Magna Græcia’ of Catanzaro, Italy. His current research interests include bioinformatics, health informatics, artificial intelligence, data mining, parallel computing. He has published six books and more than 300 papers in international journals and conference proceedings. Mario Cannataro is editor-in-chief of the Encyclopedia of Bioinformatics and Computational Biology, 2nd edn and associate editor of the Briefings in Bioinformatics and IEEE/ACM Transactions on Computational Biology and Bioinformatics journals. He is a senior member of ACM, ACM SIGBio, IEEE, IEEE Computer Society, Bioinformatics Italian Society (BITS) and Italian Society of Biomedical Informatics (SIBIM). He regularly co-organizes international workshops on bioinformatics and high-performance computing in primary conferences such as ACM-BCB, IEEE-BIBM and ICCS.

BACKGROUND: Network graphs allow modelling the real world objects in terms of interactions. In a multilayer network, the interactions are distributed over layers (i.e., intralayer and interlayer edges). Network alignment (NA) is a methodology that allows mapping nodes between two or multiple given networks, by preserving topologically similar regions. For instance, NA can be applied to transfer knowledge from one biological species to another. In this paper, we present DANTEml, a software tool for the Pairwise Global NA (PGNA) of multilayer networks, based on topological assessment. It builds its own similarity matrix by processing the node embeddings computed from two multilayer networks of interest, to evaluate their topological similarities. The proposed solution can be used via a user-friendly command line interface, also having a built-in guided mode (step-by-step) for defining input parameters. RESULTS: We investigated the performance of DANTEml based on (i) performance evaluation on synthetic multilayer networks, (ii) statistical assessment of the resulting alignments, and (iii) alignment of real multilayer networks. DANTEml over performed a method that does not consider the distribution of nodes and edges over multiple layers by 1193.62%, and a method for temporal NA by 25.88%; we also performed the statistical assessment, which corroborates the significance of its own node mappings. In addition, we tested the proposed solution by using a real multilayer network in presence of several levels of noise, in accordance with the same outcome pursued for the NA on our dataset of synthetic networks. In this case, the improvement is even more evident: +4008.75% and +111.72%, compared to a method that does not consider the distribution of nodes and edges over multiple layers and a method for temporal NA, respectively. CONCLUSIONS: DANTEml is a software tool for the PGNA of multilayer networks based on topological assessment, that is able to provide effective alignments both on synthetic and real multi layer networks, of which node mappings can be validated statistically. Our experimentation reported a high degree of reliability and effectiveness for the proposed solution.

BACKGROUND: Microarrays can perform large scale studies of differential expressed gene (DEGs) and even single nucleotide polymorphisms (SNPs), thereby screening thousands of genes for single experiment simultaneously. However, DEGs and SNPs are still just as enigmatic as the first sequence of the genome. Because they are independent from the affected biological context. Pathway enrichment analysis (PEA) can overcome this obstacle by linking both DEGs and SNPs to the affected biological pathways and consequently to the underlying biological functions and processes. RESULTS: To improve the enrichment analysis results, we present a new statistical network pre-processing method by mapping DEGs and SNPs on a biological network that can improve the relevance and significance of the DEGs or SNPs of interest to incorporate pathway topology information into the PEA. The proposed methodology improves the statistical significance of the PEA analysis in terms of computed p value for each enriched pathways and limit the number of enriched pathways. This helps reduce the number of relevant biological pathways with respect to a non-specific list of genes. CONCLUSION: The proposed method provides two-fold enhancements. Network analysis reveals fewer DEGs, by selecting only relevant DEGs and the detected DEGs improve the enriched pathways' statistical significance, rather than simply using a general list of genes.

The novel COVID-19 pandemic has posed unprecedented challenges to the society and the health sector all over the globe. Here, we present a new network-based methodology to analyze COVID-19 data measures and its application on a real dataset. The goal of the methodology is to analyze set of homogeneous datasets (i.e. COVID-19 data in several regions) using a statistical test to find similar/dissimilar dataset, mapping such similarity information on a graph and then using community detection algorithm to visualize and analyze the initial dataset. The methodology and its implementation as R function are publicly available at https://github.com/mmilano87/analyzeC19D. We evaluated diverse Italian COVID-19 data made publicly available by the Italian Protezione Civile Department at https://github.com/pcm-dpc/COVID-19/

The Disease Ontology (DO) is standardized, controlled vocabulary that contains information about inherited, developmental and acquired human diseases. Each DO term is associated with disease concepts through an annotation process. The relevance and the specificity of DO terms are often evaluated by its Information Content (IC). An important research area focus on the analysis of annotated data with the goal to extract knowledge. For example, the analysis of annotated data using Association Rules (AR) may supply meaningful knowledge, discovering relevant associations. Classical association rules methods consider all annotation equally, do not taking into account that the DO terms have different Information Content, i.e. different relevance. This implies the generation of association rules with low IC. In this paper we presents WARDO (Weighted Association Rule mining from Disease Ontology), a methodology based on the extraction od Weighted Association Rules from the DO Ontology considering the IC of terms. To assess our methodology, we tested WARDO on DO annotation datasets. WARDO is publicly available at https://gitlab.com/giuseppeagapito/wardo.

Networks are successfully used as a modelling framework in many application domains. For instance, Protein-Protein Interaction Networks (PPINs) model the set of interactions among proteins in a cell. A critical application of network analysis is the comparison among PPINs of different organisms to reveal similarities among the underlying biological processes. Algorithms for comparing networks (also referred to as network aligners) fall into two main classes: global aligners, which aim to compare two networks on a global scale, and local aligners that aim to evidence single sub-regions of similarity among networks. The possibility to improve the performance of the aligners by mixing the two approaches is a growing research area. In our previous work, we started to explore the possibility to use global alignment to improve the local one. We here explore further this possibility by using topological information extracted from global alignment to guide the steps of the local alignment. Therefore, we present Global Local Aligner (GLAlign), a methodology that improves the performances of local network aligners by exploiting a preliminary global alignment. Furthermore, we provide implementation of GLAlign. As a proof-of-principle, we evaluated the performance of the GLAlign prototype on the PPINs of five species. Results show that GLAlign methodology outperforms the state-of-the-arts local alignment algorithms. GLAlign is publicly available for academic use and can be downloaded here: https://sites.google.com/site/globallocalalignment/.

BACKGROUND: Recently the study of the complex system of connections in neural systems, i.e. the connectome, has gained a central role in neurosciences. The modeling and analysis of connectomes are therefore a growing area. Here we focus on the representation of connectomes by using graph theory formalisms. Macroscopic human brain connectomes are usually derived from neuroimages; the analyzed brains are co-registered in the image domain and brought to a common anatomical space. An atlas is then applied in order to define anatomically meaningful regions that will serve as the nodes of the network - this process is referred to as parcellation. The atlas-based parcellations present some known limitations in cases of early brain development and abnormal anatomy. Consequently, it has been recently proposed to perform atlas-free random brain parcellation into nodes and align brains in the network space instead of the anatomical image space, as a way to deal with the unknown correspondences of the parcels. Such process requires modeling of the brain using graph theory and the subsequent comparison of the structure of graphs. The latter step may be modeled as a network alignment (NA) problem. RESULTS: In this work, we first define the problem formally, then we test six existing state of the art of network aligners on diffusion MRI-derived brain networks. We compare the performances of algorithms by assessing six topological measures. We also evaluated the robustness of algorithms to alterations of the dataset. CONCLUSION: The results confirm that NA algorithms may be applied in cases of atlas-free parcellation for a fully network-driven comparison of connectomes. The analysis shows MAGNA++ is the best global alignment algorithm. The paper presented a new analysis methodology that uses network alignment for validating atlas-free parcellation brain connectomes. The methodology has been experimented on several brain datasets.

During the last years, the graph alignment has been used as a possible way to compare biological networks in system biology. The techniques for the alignment of biological networks fall into two categories: global alignment, that aims to identify large common subnetworks optimizing a topological alignment quality, and local alignment that aims to evidence single sub-regions optimizing functional alignment quality. In this work, we presented GLAlign (Global Local Aligner), a novel algorithm that integrates global and local alignment, starting from the possibility that the topological information gathered by results of global alignment can be used to improve the local alignment building. Initially, the algorithm enables the calculation of global alignment, then it uses this one to guide the building of the local alignment. GLAlign is based on two global and local algorithms widely used in literature, MAGNA++ and AlignMCL. We tested GLAlign as proof-of-principle using the Protein Interaction Networks (PINs) of three species: fly, yeast and worm. GLAlign is publicly available for academic use at https://sites.google.com/site/globallocalalignment/.

Predictive, preventive, personalized and participatory (P4) medicine is an emerging medical model that is based on the customization of all medical aspects (i.e. practices, drugs, decisions) of the individual patient. P4 medicine presupposes the elucidation of the so-called omic world, under the assumption that this knowledge may explain differences of patients with respect to disease prevention, diagnosis and therapies. Here, we elucidate the role of some selected omics sciences for different aspects of disease management, such as early diagnosis of diseases, prevention of diseases, selection of personalized appropriate and optimal therapies based on molecular profiling of patients. After introducing basic concepts of P4 medicine and omics sciences, we review some computational tools and approaches for analysing selected omics data, with a special focus on microarray and mass spectrometry data, which may be used to support P4 medicine. Some applications of biomarker discovery and pharmacogenomics and some experiences on the study of drug reactions are also described.

Gene Ontology (GO) is a structured repository of concepts (GO Terms) that are associated to one or more gene products through a process referred to as annotation. The analysis of annotated data is an important opportunity for bioinformatics. There are different approaches of analysis, among those, the use of association rules (AR) which provides useful knowledge, discovering biologically relevant associations between terms of GO, not previously known. In a previous work, we introduced GO-WAR (Gene Ontology-based Weighted Association Rules), a methodology for extracting weighted association rules from ontology-based annotated datasets. We here adapt the GO-WAR algorithm to mine cross-ontology association rules, i.e., rules that involve GO terms present in the three sub-ontologies of GO. We conduct a deep performance evaluation of GO-WAR by mining publicly available GO annotated datasets, showing how GO-WAR outperforms current state of the art approaches.

The Gene Ontology (GO) is the major resource of annotations for genes and proteins. Despite the presence of large efforts to avoid errors and inconsistencies, some unreliabilities are still present. In particular electronically inferred annotations are more unreliable than manual ones and their number is growing. Thus, the need for an accurate evaluation of annotations in an automatic way arises. In the past, some approaches for improving annotation consistencies have been proposed using association rule mining to discover hidden relationships among GO terms. However such approaches consider all the GO terms equally, while GO terms have different Information Content, i.e. different relevance. Consequently we designed a novel algorithm, (GO-WAR), Mining Weighted Association Rules from GO, that is based on the extraction of weighted association rules considering the IC of terms. We evaluated our algorithm considering seven different species and all the GO ontologies. In all the experiments GO-WAR outperformed state of the art approaches.

The Gene Ontology (GO) is used to achieve information about gene and protein functions by using a structured vocabulary of terms (GO Terms). GO Terms are related to biological concepts such as proteins or genes through the annotation process. There exist many different annotation processes identified by different evidence codes (EC). Annotated data are stored in public databases such as the Gene Ontology Annotation (GOA) database. Each term has a different specificity also referred to as Information Content (IC) of terms. Both the structure of GO and the corpora of annotation are continuously subject to change due to novel experimental findings. This process is often referred to as ontology evolution. This work focuses on how changes of annotations affect the IC of terms. The study confirms that statistically significant difference among many whole GOA versions exists on each species. Furthermore, there is also a statistically significant difference considering MF taxonomy for human, yeast, worm and fly. These results convey that annotation corpora changes have a high impact on IC.