MOTIVATION: The process of drug development is inherently complex, marked by extended intervals from the inception of a pharmaceutical agent to its eventual launch in the market. Additionally, each phase in this process is associated with a significant failure rate, amplifying the inherent challenges of this task. Computational virtual screening powered by machine learning algorithms has emerged as a promising approach for predicting therapeutic efficacy. However, the complex relationships between the features learned by these algorithms can be challenging to decipher. RESULTS: We have engineered an artificial neural network model designed specifically for predicting drug sensitivity. This model utilizes a biologically informed visible neural network, thereby enhancing its interpretability. The trained model allows for an in-depth exploration of the biological pathways integral to prediction and the chemical attributes of drugs that impact sensitivity. Our model harnesses multiomics data derived from a different tumor tissue sources, as well as molecular descriptors that encapsulate the properties of drugs. We extended the model to predict drug synergy, resulting in favorable outcomes while retaining interpretability. Given the imbalanced nature of publicly available drug screening datasets, our model demonstrated superior performance to state-of-the-art visible machine learning algorithms. AVAILABILITY AND IMPLEMENTATION: MOViDA is implemented in Python using PyTorch library and freely available for download at https://github.com/Luigi-Ferraro/MOViDA. Training data, RIS score and drug features are archived on Zenodo https://doi.org/10.5281/zenodo.8180380.

The stratification of patients at risk of progression of COVID-19 and their molecular characterization is of extreme importance to optimize treatment and to identify therapeutic options. The bioinformatics community has responded to the outbreak emergency with a set of tools and resource to identify biomarkers and drug targets that we review here. Starting from a consolidated corpus of 27 570 papers, we adopt latent Dirichlet analysis to extract relevant topics and select those associated with computational methods for biomarker identification and drug repurposing. The selected topics span from machine learning and artificial intelligence for disease characterization to vaccine development and to therapeutic target identification. Although the way to go for the ultimate defeat of the pandemic is still long, the amount of knowledge, data and tools generated so far constitutes an unprecedented example of global cooperation to this threat.

MOTIVATION: The cost of drug development has dramatically increased in the last decades, with the number new drugs approved per billion US dollars spent on R&D halving every year or less. The selection and prioritization of targets is one the most influential decisions in drug discovery. Here we present a Gaussian Process model for the prioritization of drug targets cast as a problem of learning with only positive and unlabeled examples. RESULTS: Since the absence of negative samples does not allow standard methods for automatic selection of hyperparameters, we propose a novel approach for hyperparameter selection of the kernel in One Class Gaussian Processes. We compare our methods with state-of-the-art approaches on benchmark datasets and then show its application to druggability prediction of oncology drugs. Our score reaches an AUC 0.90 on a set of clinical trial targets starting from a small training set of 102 validated oncology targets. Our score recovers the majority of known drug targets and can be used to identify novel set of proteins as drug target candidates. AVAILABILITY AND IMPLEMENTATION: The matrix of features for each protein is available at: https://bit.ly/3iLgZTa. Source code implemented in Python is freely available for download at https://github.com/AntonioDeFalco/Adaptive-OCGP. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Small non-coding RNAs (ncRNAs) are short non-coding sequences involved in gene regulation in many biological processes and diseases. The lack of a complete comprehension of their biological functionality, especially in a genome-wide scenario, has demanded new computational approaches to annotate their roles. It is widely known that secondary structure is determinant to know RNA function and machine learning based approaches have been successfully proven to predict RNA function from secondary structure information. Here we show that RNA function can be predicted with good accuracy from a lightweight representation of sequence information without the necessity of computing secondary structure features which is computationally expensive. This finding appears to go against the dogma of secondary structure being a key determinant of function in RNA. Compared to recent secondary structure based methods, the proposed solution is more robust to sequence boundary noise and reduces drastically the computational cost allowing for large data volume annotations. Scripts and datasets to reproduce the results of experiments proposed in this study are available at: https://github.com/bioinformatics-sannio/ncrna-deep.

BACKGROUND: Long non-coding RNAs (lncRNAs) represent a novel class of non-coding RNAs having a crucial role in many biological processes. The identification of long non-coding homologs among different species is essential to investigate such roles in model organisms as homologous genes tend to retain similar molecular and biological functions. Alignment-based metrics are able to effectively capture the conservation of transcribed coding sequences and then the homology of protein coding genes. However, unlike protein coding genes the poor sequence conservation of long non-coding genes makes the identification of their homologs a challenging task. RESULTS: In this study we compare alignment-based and alignment-free string similarity metrics and look at promoter regions as a possible source of conserved information. We show that promoter regions encode relevant information for the conservation of long non-coding genes across species and that such information is better captured by alignment-free metrics. We perform a genome wide test of this hypothesis in human, mouse, and zebrafish. CONCLUSIONS: The obtained results persuaded us to postulate the new hypothesis that, unlike protein coding genes, long non-coding genes tend to preserve their regulatory machinery rather than their transcribed sequence. All datasets, scripts, and the prediction tools adopted in this study are available at https://github.com/bioinformatics-sannio/lncrna-homologs .

BACKGROUND: The unveiling of long non-coding RNAs as important gene regulators in many biological contexts has increased the demand for efficient and robust computational methods to identify novel long non-coding RNAs from transcripts assembled with high throughput RNA-seq data. Several classes of sequence-based features have been proposed to distinguish between coding and non-coding transcripts. Among them, open reading frame, conservation scores, nucleotide arrangements, and RNA secondary structure have been used with success in literature to recognize intergenic long non-coding RNAs, a particular subclass of non-coding RNAs. RESULTS: In this paper we perform a systematic assessment of a wide collection of features extracted from sequence data. We use most of the features proposed in the literature, and we include, as a novel set of features, the occurrence of repeats contained in transposable elements. The aim is to detect signatures (groups of features) able to distinguish long non-coding transcripts from other classes, both protein-coding and non-coding. We evaluate different feature selection algorithms, test for signature stability, and evaluate the prediction ability of a signature with a machine learning algorithm. The study reveals different signatures in human, mouse, and zebrafish, highlighting that some features are shared among species, while others tend to be species-specific. Compared to coding potential tools and similar supervised approaches, including novel signatures, such as those identified here, in a machine learning algorithm improves the prediction performance, in terms of area under precision and recall curve, by 1 to 24%, depending on the species and on the signature. CONCLUSIONS: Understanding which features are best suited for the prediction of long non-coding RNAs allows for the development of more effective automatic annotation pipelines especially relevant for poorly annotated genomes, such as zebrafish. We provide a web tool that recognizes novel long non-coding RNAs with the obtained signatures from fasta and gtf formats. The tool is available at the following url: http://www.bioinformatics-sannio.org/software/ .

Reverse engineering of regulatory relationships from genomics data is emerging as crucial to dissect the complex underlying regulatory mechanism occurring in a cell. In this paper we propose a novel reverse engineering algorithm that makes use of formal methods, usually adopted in engineering to specify and verify concurrent software and hardware systems. With a formal specification of gene regulatory hypotheses we are able to prove mathematically whether a time course experiment belongs or not to the formal specification, determining in fact whether a gene regulation exists or not. The method is capable to detect both direction and sign (inhibition/activation) of regulations whereas most of literature methods which are limited to undirected and/or unsigned relationships. The method, empirically evaluated on experimental and synthetic datasets, reaches high levels of accuracy, outperforming literature methods in terms of precision and recall, despite the computational cost increases exponentially with the size of the network.

Online discussion forums represent, nowadays, one of the main asynchronous communication means and information sources over the Internet. The forum paradigm is adopted by the most followed websites, such as social networks and blogs. The effectiveness of discussion forums as information source, i.e., the capability to satisfy their users' information needs, depends on their information richness first, but also on how their are organized and effectively moderated. Forums organized and moderated by topics of discussion tend to host messages on related subjects and, overall, provide a classification of message threads which eases information search. In this paper we propose a semi-automatic approach to detect topics of discussion in a forum and to enhance its organization by providing a hierarchical topic-driven navigation view on its messages. We adopt Information Retrieval (IR) techniques, such as topic modeling, and formal concept analysis (FCA) to identify discussion topics and to provide a hierarchical topic-centered view on messages. We tested the validity of our approach on four forums of the e-learning platform of an Italian distance-learning university which provides around 20 moderated and unmoderated main forums followed actively by almost 5000 users, including students and teachers, each year. We validated the topics identification and messages to topics allocation process with a specific empirical experiment obtaining promising results.

Emails and issue reports capture useful knowledge about development practices, bug fixing, and change activities. Extracting such a content is challenging, due to the mix-up of source code and natural language, unstructured text.

BACKGROUND: Supervised machine learning approaches have been recently adopted in the inference of transcriptional targets from high throughput trascriptomic and proteomic data showing major improvements from with respect to the state of the art of reverse gene regulatory network methods. Beside traditional unsupervised techniques, a supervised classifier learns, from known examples, a function that is able to recognize new relationships for new data. In the context of gene regulatory inference a supervised classifier is coerced to learn from positive and unlabeled examples, as the counter negative examples are unavailable or hard to collect. Such a condition could limit the performance of the classifier especially when the amount of training examples is low. RESULTS: In this paper we improve the supervised identification of transcriptional targets by selecting reliable counter negative examples from the unlabeled set. We introduce an heuristic based on the known topology of transcriptional networks that in fact restores the conventional positive/negative training condition and shows a significant improvement of the classification performance. We empirically evaluate the proposed heuristic with the experimental datasets of Escherichia coli and show an example of application in the prediction of BCL6 direct core targets in normal germinal center human B cells obtaining a precision of 60%. CONCLUSIONS: The availability of only positive examples in learning transcriptional relationships negatively affects the performance of supervised classifiers. We show that the selection of reliable negative examples, a practice adopted in text mining approaches, improves the performance of such classifiers opening new perspectives in the identification of new transcriptional targets.

Cross-system bug fixing propagation is frequent among systems having similar characteristics, using a common framework, or, in general, systems with cloned source code fragments. While previous studies showed that clones tend to be properly maintained within a single system, very little is known about cross-system bug management.

Change impact analysis aims at identifying software artifacts being affected by a change. In the past, this problem has been addressed by approaches relying on static, dynamic, and textual analysis. Recently, techniques based on historical analysis and association rules have been explored. This paper proposes a novel change impact analysis method based on the idea that the mutual relationships between software objects can be inferred with a statistical learning approach. We use the bivariate Granger causality test, a multivariate time series forecasting approach used to verify whether past values of a time series are useful for predicting future values of another time series. Results of a preliminary study performed on the Samba daemon show that change impact relationships inferred with the Granger causality test are complementary to those inferred with association rules. This opens the road towards the development of an eclectic impact analysis approach conceived by combining different techniques.

In recent years, techniques based on association rules discovery have been extensively used to determine change-coupling relations between artifacts that often changed together. Although association rules worked well in many cases, they fail to capture logical coupling relations between artifacts modified in subsequent change sets. To overcome such a limitation, we propose the use of multivariate time series analysis and forecasting, and in particular the use of Granger causality test, to determine whether a change occurred on a software artifact was consequentially related to changes occurred on some other artifacts. Results of an empirical study performed on four Java and C open source systems show that Granger causality test is able to provide a set of change couplings complementary to association rules, and a hybrid recommender built combining recommendations from association rules and Granger causality is able to achieve a higher recall than the two single techniques.

Software systems continuously change for various reasons, such as adding new features, performing bug fixing, or doing some refactoring activities. Such changes may either increase the source code complexity and disorganization, or help to reduce it. Developers apply adequate design principles and assets, including design patterns, to make software resilient to changes and control complexity. This paper empirically investigates the relationship of source code complexity and disorganization-measured using source code entropy-with three factors: different kinds of changes occurring to software systems, the presence of design patterns in the source code, and the number of contributors that modified the source code file. Results of an exploratory study carried out on an interval of the life-time span of two open source systems, ArgoUML and Eclipse-JDT, suggest that (i) different kinds of changes-namely refactorings and other kinds of changes-may contribute either negatively or positively to the entropy, (ii) the use of design patterns does not necessarily help to mitigate code degradation-thus confirming previous findings on the role played by design patterns-and (iii) entropy tends to increase with the number of file committers.

MOTIVATION: Genomic copy number (CN) information is useful to study genetic traits of many diseases. Using array comparative genomic hybridization (aCGH), researchers are able to measure the copy number of thousands of DNA loci at the same time. Therefore, a current challenge in bioinformatics is the development of efficient algorithms to detect the map of aberrant chromosomal regions. METHODS: We describe an approach for the segmentation of copy number aCGH data. Variational estimator for genomic aberrations (VEGA) adopt a variational model used in image segmentation. The optimal segmentation is modeled as the minimum of an energy functional encompassing both the quality of interpolation of the data and the complexity of the solution measured by the length of the boundaries between segmented regions. This solution is obtained by a region growing process where the stop condition is completely data driven. RESULTS: VEGA is compared with three algorithms that represent the state of the art in CN segmentation. Performance assessment is made both on synthetic and real data. Synthetic data simulate different noise conditions. Results on these data show the robustness with respect to noise of variational models and the accuracy of VEGA in terms of recall and precision. Eight mantle cell lymphoma cell lines and two samples of glioblastoma multiforme are used to evaluate the behavior of VEGA on real biological data. Comparison between results and current biological knowledge shows the ability of the proposed method in detecting known chromosomal aberrations. AVAILABILITY: VEGA has been implemented in R and is available at the address http://www.dsba.unisannio.it/Members/ceccarelli/vega in the section Download.

BACKGROUND: Recently, supervised learning methods have been exploited to reconstruct gene regulatory networks from gene expression data. The reconstruction of a network is modeled as a binary classification problem for each pair of genes. A statistical classifier is trained to recognize the relationships between the activation profiles of gene pairs. This approach has been proven to outperform previous unsupervised methods. However, the supervised approach raises open questions. In particular, although known regulatory connections can safely be assumed to be positive training examples, obtaining negative examples is not straightforward, because definite knowledge is typically not available that a given pair of genes do not interact. RESULTS: A recent advance in research on data mining is a method capable of learning a classifier from only positive and unlabeled examples, that does not need labeled negative examples. Applied to the reconstruction of gene regulatory networks, we show that this method significantly outperforms the current state of the art of machine learning methods. We assess the new method using both simulated and experimental data, and obtain major performance improvement. CONCLUSIONS: Compared to unsupervised methods for gene network inference, supervised methods are potentially more accurate, but for training they need a complete set of known regulatory connections. A supervised method that can be trained using only positive and unlabeled data, as presented in this paper, is especially beneficial for the task of inferring gene regulatory networks, because only an incomplete set of known regulatory connections is available in public databases such as RegulonDB, TRRD, KEGG, Transfac, and IPA.

Differencing tools are highly relevant for a series of software engineering tasks, including analyzing developers' activities, assessing the changeability of software artifacts, and monitoring the maintenance of critical assets such as source clones and vulnerable instructions. This tool demonstration shows the features of ldiff, an enhanced, language-independent line differencing tool. L-diff builds upon the Unix diff and overcomes its limitations in determining whether an artifact line has been changed or is the result of additions and removals, and in tracking artifact fragments that have been moved upward or downward within the file. The paper describes the tool and shows its capability of analyzing changes on different kinds of software artifacts, including use cases, code developed with different programming languages, and test cases.

Analyzing the change-proneness of design patterns and the kinds of changes occurring to classes playing role(s) in some design pattern(s) during software evolution poses the basis for guidelines to help developers who have to choose, apply or maintain design patterns. Building on previous work, this paper shifts the focus from design patterns as wholes to the finer-grain level of design pattern roles. The paper presents an empirical study to understand whether there are roles that are more change-prone than others and whether there are changes that are more likely to occur to certain roles. The study relies on data extracted from the source code repositories of three different systems (JHotDraw, Xerces, and Eclipse-JDT) and from 12 design patterns. Results obtained confirm the intuitive behavior about changeability of many roles in design motifs, but also warns about properly designing parts of the motif subject to frequent changes.

The presence of vulnerable statements in the source code is a crucial problem for maintainers: properly monitoring and, if necessary, removing them is highly desirable to ensure high security and reliability. To this aim, a number of static analysis tools have been developed to detect the presence of instructions that can be subject to vulnerability attacks, ranging from buffer overflow exploitations to command injection and cross-site scripting.Based on the availability of existing tools and of data extracted from software repositories, this paper reports an empirical study on the evolution of vulnerable statements detected in three software systems with different static analysis tools. Specifically, the study investigates on vulnerability evolution trends and on the decay time exhibited by different kinds of vulnerabilities.

Design patterns are solutions to recurring design problems, conceived to increase benefits in terms of reuse, code quality and, above all, maintainability and resilence to changes. This paper presents results from an empirical study aimed at understanding the evolution of design patterns in three open source systems, namely JHotDraw, ArgoUML, and Eclipse-JDT. Specifically, the study analyzes how frequently patterns are modified, to what changes they undergo and what classes co-change with the patterns. Results show how patterns more suited to support the application purpose tend to change more frequently, and that different kind of changes have a different impact on co-changed classes and a different capability of making the system resilent to changes.

Crosscutting concerns are software system features whose implementation is spread across many modules as tangled and scattered code. Identifying such code helps developers to change the concern and/or re-factor it to an aspect. This paper evaluates the suitability of line co-change as a technique for the identification of crosscutting concerns. A line co-change aim at identifying source code lines that have been changed together in a commit transaction performed using a versioning system such as CVS. Promising results have been obtained by evaluating the approach to identify four crosscutting concerns present in an open-source system, JHotDraw. The paper also shows that line co-change can be effectively complemented with clone detection to improve the performance achieved by the separate approaches

The assessment of software systems often requires to consider together qualitative and quantitative aspects. Because of the different nature, measures belong to different domains. The main problem is to aggregate such information into a derived measure able to provide an overall estimation. This problem has been traditionally solved trough the transformation of qualitative assessments into quantitative measures. Indeed, such a transformation implicitly assumes a conceptual equivalence between the terms quantitative and objective on one side, and qualitative and subjective on the other side. An alternative approach is to consider logical aggregation models, able to infer the overall evaluation based on the assessment of individual attributes. This approach requires an early transformation of quantitative measures in qualitative assessments. Such a transformation is possible trough the use of judgment functions. The aim of this paper is to introduce the judgment functions and to study their properties.

XML is the most influential standard for data exchange and Web presentation. The power of XML derives from the fact that the aspects of structuring, representing and visualizing a piece of information are handled independently with specific tools.In this article we deal with the information visualization aspect, which in XML is managed through XSLT transformations, and propose a visual approach to define XML to FO transformations. The approach has been implemented in a graphical environment that eases the definition of the format and visualization of XML documents by means of a graphicalcontent/container metaphor.