Machine learning is a powerful approach for analysing RNA sequences, particularly for understanding the function and regulation of noncoding RNAs. A critical step in this process is feature extraction, which transforms biological sequences into numerical representations that allow computational models to capture and interpret complex biological patterns. Despite its central role, the field of RNA feature extraction remains broad and fragmented, with limited standardization and accessibility hindering consistent application. In this comprehensive review, we address the fragmentation of the field by systematically organizing over 25 feature extraction strategies into sequence- and structure-based approaches. We further conduct a comparative analysis highlighting how the choice of feature sets impacts model performance, reinforcing the importance of integrated feature engineering. To facilitate practical adoption, it also provides a curated list of publicly available tools and software packages. By consolidating methodologies and resources, this work seeks to improve reproducibility, scalability, and interpretability in machine learning-driven RNA research.

The complementary information found in different modalities of patient data can aid in more accurate modelling of a patient's disease state and a better understanding of the underlying biological processes of a disease. However, the analysis of multi-modal, multi-omics data presents many challenges. In this work, we compare the performance of a variety of ensemble machine learning (ML) algorithms that are capable of late integration of multi-class data from different modalities. The ensemble methods and their variations tested were (i) a voting ensemble, with hard and soft vote, (ii) a meta learner, and (iii) a multi-modal AdaBoost model using hard vote, soft vote, and meta learner to integrate the modalities on each boosting round, the PB-MVBoost model and a novel application of a mixture of expert's model. These were compared to simple concatenation. We examine these methods using data from an in-house study on hepatocellular carcinoma, plus validation datasets on studies from breast cancer and irritable bowel disease. We develop models that achieve an area under the receiver operating curve of up to 0.85 and find that two boosted methods, PB-MVBoost and AdaBoost with soft vote were the best performing models. We also examine the stability of features selected and the size of the clinical signature. Our work shows that integrating complementary omics and data modalities with effective ensemble ML models enhances accuracy in multi-class clinical outcome predictions and produces more stable predictive features than individual modalities or simple concatenation. We provide recommendations for the integration of multi-modal multi-class data.

Gene expression regulation is a sophisticated, multi-stage process, and its robustness is critical to normal cell function and the survival of an organism. Previous studies indicate that differential gene expression at the RNA level is typically attenuated at the protein level through translational regulation. However, how post-transcriptional regulation (PTR) influences expression change during the RNA maturation process remains unclear. In this study, we investigated this by quantifying the magnitude of expression change in precursor RNA and mature RNA across a vast range of different biological conditions. We analyzed bulk tissue RNA sequencing data from 4689 samples, including healthy and diseased tissues from human, chimpanzee, rhesus macaque, and murine sources. We demonstrated that PTR tends to support homeostatic expression of mature RNA by amplifying normal tissue-specific expression of precursor RNA, while reducing expression change of precursor RNA in disease contexts. Our study provides insight into the general influence of PTR on gene expression homeostasis. Our analysis also suggests that intronic reads in RNA-seq studies may contain under-utilized information about disease associations. Additionally, our findings may assist in identifying new disease biomarkers and more effective ways of altering gene expression as a therapeutic strategy.

Cancer begins when healthy cells change and grow out of control, forming a mass called a tumor. Head and neck (H&N) cancers usually develop in or around the head and neck, including the mouth (oral cavity), nose and sinuses, throat (pharynx), and voice box (larynx). 4% of all cancers are H&N cancers with a very low survival rate (a five-year survival rate of 64.7%). FDG-PET/CT imaging is often used for early diagnosis and staging of H&N tumors, thus improving these patients' survival rates. This work presents a novel 3D-Inception-Residual aided with 3D depth-wise convolution and squeeze and excitation block. We introduce a 3D depth-wise convolution-inception encoder consisting of an additional 3D squeeze and excitation block and a 3D depth-wise convolution-based residual learning decoder (3D-IncNet), which not only helps to recalibrate the channel-wise features but adaptively through explicit inter-dependencies modeling but also integrate the coarse and fine features resulting in accurate tumor segmentation. We further demonstrate the effectiveness of inception-residual encoder-decoder architecture in achieving better dice scores and the impact of depth-wise convolution in lowering the computational cost. We applied random forest for survival prediction on deep, clinical, and radiomics features. Experiments are conducted on the benchmark HECKTOR21 challenge, which showed significantly better performance by surpassing the state-of-the-artwork and achieved 0.836 and 0.811 concordance index and dice scores, respectively. We made the model and code publicly available.

Liquid biopsy has shown promise for cancer diagnosis due to its minimally invasive nature and the potential for novel biomarker discovery. However, the low concentration of relevant blood-based biosources and the heterogeneity of samples (i.e. the variability of relative abundance of molecules identified), pose major challenges to biomarker discovery. Moreover, the number of molecular measurements or features (e.g. transcript read counts) per sample could be in the order of several thousand, whereas the number of samples is often substantially lower, leading to the curse of dimensionality. These challenges, among others, elucidate the importance of a robust biomarker panel identification or feature extraction step wherein relevant molecular measurements are identified prior to classification for cancer detection. In this work, we performed a benchmarking study on 12 feature extraction methods using transcriptomic profiles derived from different blood-based biosources. The methods were assessed both in terms of their predictive performance and the robustness of the biomarker panels in diagnosing cancer or stratifying cancer subtypes. While performing the comparison, the feature extraction methods are categorized into feature subset selection methods and transformation methods. A transformation feature extraction method, namely partial least square discriminant analysis, was found to perform consistently superior in terms of classification performance. As part of the benchmarking study, a generic pipeline has been created and made available as an R package to ensure reproducibility of the results and allow for easy extension of this study to other datasets (https://github.com/VafaeeLab/bloodbased-pancancer-diagnosis).

To accurately predict the regional spread of coronavirus disease 2019 (COVID-19) infection, this study proposes a novel hybrid model, which combines a long short-term memory (LSTM) artificial recurrent neural network with dynamic behavioral models. Several factors and control strategies affect the virus spread, and the uncertainty arising from confounding variables underlying the spread of the COVID-19 infection is substantial. The proposed model considers the effect of multiple factors to enhance the accuracy in predicting the number of cases and deaths across the top ten most-affected countries at the time of the study. The results show that the proposed model closely replicates the test data, such that not only it provides accurate predictions but it also replicates the daily behavior of the system under uncertainty. The hybrid model outperforms the LSTM model while accounting for data limitation. The parameters of the hybrid models are optimized using a genetic algorithm for each country to improve the prediction power while considering regional properties. Since the proposed model can accurately predict the short-term to medium-term daily spreading of the COVID-19 infection, it is capable of being used for policy assessment, planning, and decision making.

Drug similarity studies are driven by the hypothesis that similar drugs should display similar therapeutic actions and thus can potentially treat a similar constellation of diseases. Drug-drug similarity has been derived by variety of direct and indirect sources of evidence and frequently shown high predictive power in discovering validated repositioning candidates as well as other in-silico drug development applications. Yet, existing resources either have limited coverage or rely on an individual source of evidence, overlooking the wealth and diversity of drug-related data sources. Hence, there has been an unmet need for a comprehensive resource integrating diverse drug-related information to derive multi-evidenced drug-drug similarities. We addressed this resource gap by compiling heterogenous information for an exhaustive set of small-molecule drugs (total of 10 367 in the current version) and systematically integrated multiple sources of evidence to derive a multi-modal drug-drug similarity network. The resulting database, 'DrugSimDB' currently includes 238 635 drug pairs with significant aggregated similarity, complemented with an interactive user-friendly web interface (http://vafaeelab.com/drugSimDB.html), which not only enables database ease of access, search, filtration and export, but also provides a variety of complementary information on queried drugs and interactions. The integration approach can flexibly incorporate further drug information into the similarity network, providing an easily extendable platform. The database compilation and construction source-code has been well-documented and semi-automated for any-time upgrade to account for new drugs and up-to-date drug information.

A typical single-cell RNA sequencing (scRNA-seq) experiment will measure on the order of 20 000 transcripts and thousands, if not millions, of cells. The high dimensionality of such data presents serious complications for traditional data analysis methods and, as such, methods to reduce dimensionality play an integral role in many analysis pipelines. However, few studies have benchmarked the performance of these methods on scRNA-seq data, with existing comparisons assessing performance via downstream analysis accuracy measures, which may confound the interpretation of their results. Here, we present the most comprehensive benchmark of dimensionality reduction methods in scRNA-seq data to date, utilizing over 300 000 compute hours to assess the performance of over 25 000 low-dimension embeddings across 33 dimensionality reduction methods and 55 scRNA-seq datasets. We employ a simple, yet novel, approach, which does not rely on the results of downstream analyses. Internal validation measures (IVMs), traditionally used as an unsupervised method to assess clustering performance, are repurposed to measure how well-formed biological clusters are after dimensionality reduction. Performance was further evaluated over nearly 200 000 000 iterations of DBSCAN, a density-based clustering algorithm, showing that hyperparameter optimization using IVMs as the objective function leads to near-optimal clustering. Methods were also assessed on the extent to which they preserve the global structure of the data, and on their computational memory and time requirements across a large range of sample sizes. Our comprehensive benchmarking analysis provides a valuable resource for researchers and aims to guide best practice for dimensionality reduction in scRNA-seq analyses, and we highlight Latent Dirichlet Allocation and Potential of Heat-diffusion for Affinity-based Transition Embedding as high-performing algorithms.

SUMMARY: TDAview is an online tool for topological data analysis (TDA) and visualization. It implements the Mapper algorithm for TDA and provides extensive graph visualization options. TDAview is a user-friendly tool that allows biologists and clinicians without programming knowledge to harness the power of TDA. TDAview supports an analysis and visualization mode in which a Mapper graph is constructed based on user-specified parameters, followed by graph visualization. It can also be used in a visualization only mode in which TDAview is used for visualizing the data properties of a Mapper graph generated using other open-source software. The graph visualization options allow data exploration by graphical display of metadata variable values for nodes and edges, as well as the generation of publishable figures. TDAview can handle large datasets, with tens of thousands of data points, and thus has a wide range of applications for high-dimensional data, including the construction of topology-based gene co-expression networks. AVAILABILITY AND IMPLEMENTATION: TDAview is a free online tool available at https://voineagulab.github.io/TDAview/. The source code, usage documentation and example data are available at TDAview GitHub repository: https://github.com/Voineagulab/TDAview.

Bayesian Network (BN) structure learning is a complex search problem, generally characterized by multimodality and epistasis. Genetic Algorithms (GAs) have been extensively used to pursue the BN structure learning task. This paper presents a new approach which incorporates the structural properties of the problem into GA mechanisms. The proposed approach uses a new recombination operator named Parent Set crossover, capable of reducing the disruptive action of the recombination process and enhancing its exploitative power. The new operator has been compared with a comprehensive set of other crossover operators as part of two genetic strategies: a canonical GA and a GA with an adaptive mutation scheme. All examined crossover operators were applied on both canonical and adaptive GAs and then compared in terms of various performance metrics. The experiments involve performance measures at the end of evolution as well as their convergence behavior across generations. The performance of the proposed method was also compared with the state-of-the-art non-evolutionary BN structure learning algorithms. Results show that the proposed recombination method enhances the algorithmic efficiency over a variety of test cases of different size.

Table of contents A1 Highlights from the eleventh ISCB Student Council Symposium 2015 Katie Wilkins, Mehedi Hassan, Margherita Francescatto, Jakob Jespersen, R. Gonzalo Parra, Bart Cuypers, Dan DeBlasio, Alexander Junge, Anupama Jigisha, Farzana Rahman O1 Prioritizing a drug’s targets using both gene expression and structural similarity Griet Laenen, Sander Willems, Lieven Thorrez, Yves Moreau O2 Organism specific protein-RNA recognition: A computational analysis of protein-RNA complex structures from different organisms Nagarajan Raju, Sonia Pankaj Chothani, C. Ramakrishnan, Masakazu Sekijima; M. Michael Gromiha O3 Detection of Heterogeneity in Single Particle Tracking Trajectories Paddy J Slator, Nigel J Burroughs O4 3D-NOME: 3D NucleOme Multiscale Engine for data-driven modeling of three-dimensional genome architecture Przemysław Szałaj, Zhonghui Tang, Paul Michalski, Oskar Luo, Xingwang Li, Yijun Ruan, Dariusz Plewczynski O5 A novel feature selection method to extract multiple adjacent solutions for viral genomic sequences classification Giulia Fiscon, Emanuel Weitschek, Massimo Ciccozzi, Paola Bertolazzi, Giovanni Felici O6 A Systems Biology Compendium for Leishmania donovani Bart Cuypers, Pieter Meysman, Manu Vanaerschot, Maya Berg, Hideo Imamura, Jean-Claude Dujardin, Kris Laukens O7 Unravelling signal coordination from large scale phosphorylation kinetic data Westa Domanova, James R. Krycer, Rima Chaudhuri, Pengyi Yang, Fatemeh Vafaee, Daniel J. Fazakerley, Sean J. Humphrey, David E. James, Zdenka Kuncic

Exploration and exploitation are the two cornerstones which characterize Evolutionary Algorithms (EAs) capabilities. Maintaining the reciprocal balance of the explorative and exploitative power is the key to the success of EA applications. Accordingly, this work is concerned with proposing a diversity-guided genetic algorithm with a new mutation scheme that is capable of exploring the unseen regions of the search space, as well as exploiting the already-found promising elements. The proposed mutation operator specifies different mutation rates for different sites of an encoded solution. These site-specific rates are carefully derived based on the underlying pattern of highly-fit solutions, adjusted to every single individual, and adapted throughout the evolution to retain a good ratio between exploration and exploitation. Furthermore, in order to more directly monitor the exploration vs. exploitation balance, the proposed method is augmented with a diversity control process assuring that the search process does not lose the required balance between the two forces.

Bayesian networks are probabilistic graphical models representing conditional dependencies among a set of random variables. Due to their concise representation of the joint probability distribution, Bayesian Networks are becoming incrementally popular models for knowledge representation and reasoning in various problem domains. However, learning the structure of the Bayesian networks is an NP-hard problem since the number of structures grows super-exponentially as the number of variables increases. This work therefore is aimed to propose a new hybrid structure learning algorithm that uses mutual dependencies to reduce the search space complexity and recruits the genetic algorithm to effectively search over the reduced space of possible structures. The proposed method is best suited for problems with medium to large number of variables and a limited dataset. It is shown that the proposed method achieves higher model's accuracy as compared to a series of popular structure learning algorithms particularly when the data size gets smaller.

This paper is concerned with proposing an elitist genetic algorithm which makes use of a new mutation scheme aimed to tackle both explorative and exploitative responsibilities of genetic operators. The proposed mutation scheme follows an approach similar to motif representation in biology, to derive the underlying pattern of highly-fit solutions discovered so far. This pattern is then used to derive mutation rates specified for every site along the encoded solutions. The site-specific rates are amended for every individual to balance the required explorative and exploitative power. The Markov chain model of the proposed method is also derived and used to analyze its convergence properties.

Exploration and exploitation are the two cornerstones which characterize Evolutionary Algorithms (EAs) capabilities. Maintaining the reciprocal balance of the explorative and exploitative power is the key to the success of EA applications. Accordingly, in this work the canonical Genetic Algorithm is augmented by a new mutation scheme that is capable of exploring the unseen regions of the search space, and simultaneously exploiting the already-found promising elements. The proposed mutation operator specifies different mutation rates for different sites (loci) of the individuals. These site-specific rates are wisely derived based on the fitness and structure of the population individuals. In order to retain the balance of the required exploration and exploitation, the mutation rates are adapted during the evolution. To demonstrate the efficacy of the proposed algorithm, the method is evaluated using a set of benchmark problems and the outcome is compared with a series of well-known relevant algorithms. The results demonstrate that the newly suggested method significantly outperforms its rivals.

This work is concerned with proposing a robust framework for optimizing operator rates of simple Genetic Algorithms (GAs) during a GA run. The suggested framework is built upon a formerly proposed GA Markov chain model to estimate the optimal values of the operator rates based on the time and the current state of the evolution. Though the proposed framework has been formalized for optimizing both mutation and crossover rates, in the current paper, we only implemented it as the mutation rate optimizer and kept the crossover rate constant. To demonstrate the efficacy of the proposed algorithm, the method is evaluated using a set of benchmark problems and the outcome is compared with a series of well-known relevant algorithms. The results demonstrate that the newly suggested algorithm significantly outperforms its rivals.

The in-network query processing paradigm in sensor networks postulates that a query is routed among sensors and collects the answers from the sensors on its trajectory. It works for static and connected sensor networks. However, when the network consists of mobile sensors and is sparse, a different approach is necessary. In this paper we propose a query processing method that uses cooperative caching. It makes the data items satisfying a query flow to its originator. To cope with communication bandwidth and storage constraints, the method prioritizes the data-items in terms of their value, as reflected by supply and demand. Simulations based on real-life mobility traces identify the situations in which our approach outperforms a series of existing cooperative caching strategies and an existing mobile sensor network algorithm.

Dealing with many free parameters and finding an appropriate set of parameter values for an evolutionary algorithm (EA) has been a longstanding major challenge of the evolutionary computation community. Such difficulty has directed researchers' attention towards devising an automated ways of controlling EA parameters. This work is concerned with proposing a novel method which adaptively adjusts EA and specifically genetic algorithm (GA) mutation rates. The proposed method incorporates the underlying statistical framework of biological evolutionary models into the generic context of evolutionary algorithms. By using such model, besides adapting the mutation rate, this method aims to wisely determine the types of replacing genes in the mutation procedure. To demonstrate the efficacy of the proposed algorithm, the method is evaluated using a wide array of test functions and the outcome is compared with a state-of-the-art adaptive mutation evolutionary algorithm. The results demonstrates that the newly suggested algorithm significantly outperform its adaptive rival in most of the test cases.

This work is concerned with proposing an adaptive method to dynamically adjust genetic operator probabilities throughout the evolutionary process. The proposed method relies on the individual preferences of each chromosome, rather than the global behavior of the whole population. Hence, each individual carries its own set of parameters, including the probabilities of the genetic operators. The carried parameters undergo the same evolutionary process as the carriers--the chromosomes - do. We call this method Evolved Evolutionary Algorithm (E2A) as it has an additional evolutionary process to evolve control parameters. Furthermore, E2A employs a supplementary mutation operator (DE-mutation) which utilizes the previously overlooked numerical optimization model known as the Differential Evolution to expedite the optimization rate of the genetic parameters. To leverage our previous work, we used Gene Expression Programming (GEP) as a benchmark to determine the performance of our proposed method. Nevertheless, E2A can be easily extended to other genetic programming variants. As the experimental results on a wide array of regression problems demonstrate, the E2A method reveals a faster rate of convergence and provides fitter ultimate solutions. However, to further expose the power of the E2A method, we compared it to related methods using self-adaptation previously applied to Genetic Algorithms. Our benchmarking on the same set of regression problems proves the supremacy of our proposed method both in the accuracy and simplicity of the final solutions.