Volume 16, Issue 1 - Special Issue: Selected extended papers of the 12th International Conference on Practical Applications of Computational Biology and Bioinformatics (PACBB) / Guest Editors: Florentino Fdez-Riverola, Miguel Rocha

The advance of high-throughput RNA-Sequencing techniques enables researchers to analyze the complete gene activity in particular cells. From the insights of such analyses, researchers can identify disease-specific expression profiles, thus understand complex diseases like cancer, and eventually develop effective measures for diagnosis and treatment. The high dimensionality of gene expression data poses challenges to its computational analysis, which is addressed with measures of gene selection. Traditional gene selection approaches base their findings on statistical analyses of the actual expression levels, which implies several drawbacks when it comes to accurately identifying the underlying biological processes. In turn, integrative approaches include curated information on biological processes from external knowledge bases during gene selection, which promises to lead to better interpretability and improved predictive performance. Our work compares the performance of traditional and integrative gene selection approaches. Moreover, we propose a straightforward approach to integrate external knowledge with traditional gene selection approaches. We introduce a framework enabling the automatic external knowledge integration, gene selection, and evaluation. Evaluation results prove our framework to be a useful tool for evaluation and show that integration of external knowledge improves overall analysis results.

Parkinson’s disease is one of the most common neurodegenerative illnesses in older persons and the leucine-rich repeat kinase 2 (LRRK2) is an auspicious target for its pharmacological treatment. In this work, quantitative structure–activity relationship (QSAR) models for identification of putative inhibitors of LRRK2 protein are developed by using an in-house chemical library and several machine learning techniques. The methodology applied in this paper has two steps: first, alternative subsets of molecular descriptors useful for characterizing LRRK2 inhibitors are chosen by a multi-objective feature selection method; secondly, QSAR models are learned by using these subsets and three different strategies for supervised learning. The qualities of all these QSAR models are compared by classical metrics and the best models are discussed in statistical and physicochemical terms.

We present a flexible deep convolutional neural network method for the analysis of arbitrary sized graph structures representing molecules. This method, which makes use of the Lipinski RDKit module, an open-source cheminformatics software, enables the incorporation of any global molecular (such as molecular charge and molecular weight) and local (such as atom hybridization and bond orders) information. In this paper, we show that this method significantly outperforms another recently proposed method based on deep convolutional neural networks on several datasets that are studied. Several best practices for training deep convolutional neural networks on chemical datasets are also highlighted within the article, such as how to select the information to be included in the model, how to prevent overfitting and how unbalanced classes in the data can be handled.

Metabolism has been a major field of study in the last years, mainly due to its importance in understanding cell physiology and certain disease phenotypes due to its deregulation. Genome-scale metabolic models (GSMMs) have been established as important tools to help achieve a better understanding of human metabolism. Towards this aim, advances in systems biology and bioinformatics have allowed the reconstruction of several human GSMMs, although some limitations and challenges remain, such as the lack of external identifiers for both metabolites and reactions. A pipeline was developed to integrate multiple GSMMs, starting by retrieving information from the main human GSMMs and evaluating the presence of external database identifiers and annotations for both metabolites and reactions. Information from metabolites was included into a graph database with omics data repositories, allowing clustering of metabolites through their similarity regarding database cross-referencing. Metabolite annotation of several older GSMMs was enriched, allowing the identification and integration of common entities. Using this information, as well as other metrics, we successfully integrated reactions from these models. These methods can be leveraged towards the creation of a unified consensus model of human metabolism.

In this paper, we propose a parallel cellular automaton tumor growth model that includes load balancing of cells distribution among computational threads with the introduction of adjusting parameters. The obtained results show a fair reduction in execution time and improved speedup compared with the sequential tumor growth simulation program currently referenced in tumoral biology. The dynamic data structures of the model can be extended to address additional tumor growth characteristics such as angiogenesis and nutrient intake dependencies.