Search

Relevant research has been highlighted in the computing community to develop machine learning models capable of predicting the occurrence of crimes, analyzing contexts of crimes, extracting profiles of individuals linked to crime, and analyzing crimes over time. However, models capable of predicting specific crimes, such as homicide, are not commonly found in the current literature. This research presents a machine learning model to predict homicide crimes, using a dataset that uses generic data (without study location dependencies) based on incident report records for 34 different types of crimes, along with time and space data from crime reports. Experimentally, data from the city of Belem - Para, Brazil was used. These data were transformed to make the problem generic, enabling the replication of this model to other locations. In the research, analyses were performed with simple and robust algorithms on the created dataset. With this, statistical tests were performed with 11 different classification methods and the results are related to the prediction’s occurrence and non-occurrence of homicide crimes in the month subsequent to the occurrence of other registered crimes, with 76% assertiveness for both classes of the problem, using Random Forest. Results are considered as a baseline for the proposed problem.

Housekeeping genes (HKGs), are essential for gene expression based studies performed through Reverse Transcriptase-polymerase Chain Reaction (RT-qPCR). These genes are related with the basic cellular processes that are essential for cell maintenance, survival and function. Thus, HKGs should be expressed in all cells of an organism regardless of the tissue type, cell state or cell condition. High-throughput technologies, including RNA sequencing (RNA-seq), are used to study and identify these types of genes. RNA-seq is a high-throughput method that allows the measurement of gene expression profiles in a target tissue or an isolated cell. Moreover, machine learning methods are routinely applied in different genomics related areas to enable the interpretation of large datasets, including those related to gene expression. This study reports a new machine learning based approach to identify candidate HKGs in silico from RNA-seq gene expression data. The approach enabled the identification of stable HKGs candidates in RNA-seq data from Corynebacterium pseudotuberculosis. These genes showed stable expression under different stress conditions as well as low variation index and fold changes. Furthermore, some of these genes were already reported in the literature as HKGs or HKGs candidates for the same or other bacterial organisms, which reinforced the accuracy of the proposed method. We present a novel approach based on K-means algorithm, internal metrics and machine learning methods that can identify stable housekeeping genes from gene expression data with high accuracy and efficiency.

In clustering, one of the most challenging aspects is the validation, whose objective is to evaluate how good a clustering solution is. Sequence binning is a clustering task on metagenomic data analysis. The sequence clustering challenge is essentially putting together sequences belonging to the same genome. As a clustering problem it requires proper use of validation criteria of the discovered partitions. In sequence binning, the concepts of precision and recall, and F-measure index (external validation) are normally used as benchmark. However, on practice, information about the (sub) optimal number of cluster is unknown, so these metrics might be biased to an overestimated “ground truth”. In the case of sequence binning analysis, where the reference information about genomes is not available, how to evaluate the quality of bins resulting from a clustering solution? To answer this question we empirically study both quantitative (internal indexes) and qualitative aspects (biological soundness) while evaluating clustering solutions on the sequence binning problem. Our experimental study indicates that the number of clusters, estimated by binning algorithms, do not have as much impact on the quality of bins by means of biological soundness of the discovered clusters. The quality of the sub-optimal bins (greater than 90%) were identified in both rich and poor clustering partitions. Qualitative validation is essential for proper evaluation of a sequence binning solution, generating bins with sub-optimal quality. Internal indexes can only be used in compliance with qualitative ones as a trade-off between the number of partitions and biological soundness of its respective bins.

Despite the availability of benchmark machine learning (ML) repositories (e.g., UCI, OpenML), there is no standard evaluation strategy yet capable of pointing out which is the best set of datasets to serve as gold standard to test different ML algorithms. In recent studies, Item Response Theory (IRT) has emerged as a new approach to elucidate what should be a good ML benchmark. This work applied IRT to explore the well-known OpenML-CC18 benchmark to identify how suitable it is on the evaluation of classifiers. Several classifiers ranging from classical to ensembles ones were evaluated using IRT models, which could simultaneously estimate dataset difficulty and classifiers’ ability. The Glicko-2 rating system was applied on the top of IRT to summarize the innate ability and aptitude of classifiers. It was observed that not all datasets from OpenML-CC18 are really useful to evaluate classifiers. Most datasets evaluated in this work (84%) contain easy instances in general (e.g., around 10% of difficult instances only). Also, 80% of the instances in half of this benchmark are very discriminating ones, which can be of great use for pairwise algorithm comparison, but not useful to push classifiers abilities. This paper presents this new evaluation methodology based on IRT as well as the tool decodIRT, developed to guide IRT estimation over ML benchmarks.

Reinforcement learning (RL) aims to build intelligent agents able to optimally act after the training process to solve a given goal task in an autonomous and non-deterministic fashion. It has been successfully employed in several areas; however, few RL-based approaches related to genome assembly have been found, especially when considering real input datasets. De novo genome assembly is a crucial step in a number of genome projects, but due to its high complexity, the outcome of state-of-art assemblers is still insufficient to assist researchers in answering all their scientific questions properly. Hence, the development of better assembler is desirable and perhaps necessary, and preliminary studies suggest that RL has the potential to solve this computational task. In this sense, this paper presents an empirical analysis to evaluate this hypothesis, particularly in higher scale, through performance assessment along with time and space complexity analysis of a theoretical approach to the problem of assembly proposed by [2] using the RL algorithm Q-learning. Our analysis shows that, although space and time complexities are limiting scale issues, RL is shown as a viable alternative for solving the DNA fragment assembly problem.

Microorganisms abound everywhere. Though we know they play key roles in several ecosystems, too little is known about how these complex communities work. To act as a community they must interact with each other in order to achieve such community stability in which proper functions allows the microbial community to adapt in complex environment conditions. Thus, to effectively understand microbial genetic networks one needs to explore them by means of a systems biology approach. The proposed approach extends the metagenomic gene-centric view by taking into account the set of genes present in a metagenome and also the complex links of interactions among these genes and by treating the microbiome as a single biological system. In this paper, we present the FUNN-MG computational framework to explore functional modules in microbial genetic networks.

The catalogs of molecular biology databases does not provide a full description of databases, so the user should select databases using limited information available. Taking into account this fact, in the context of an initiative called BioDBCore, a group of experts proposes core metadata definitions to describe the molecular biology databases. However, how to use these metadata to infer the quality of a database is a clear open issue. In the present work, we propose an ontology-based approach aiming to guide the database selection process from molecular biology database catalogs using these metadata.

Metagenomics is an emerging field in which the power of genome analysis is applied to entire communities of microbes. A large variety of classifiers has been developed for gene prediction though there is lack of an empirical evaluation regarding the core machine learning techniques implemented in these tools. In this work we present an empirical performance evaluation of classification strategies for metagenomic gene prediction. This comparison takes into account distinct supervised learning strategies: one lazy learner, two eager-learners and one ensemble learner. Though the performance of the four base classifiers was good, the ensemble-based strategy with Random Forest has achieved the overall best result.

We propose an integrative network-based meta-analysis strategy to enable the selection of potential gene markers for one of the most prevalent diseases worldwide, Type 2 diabetes (T2D), formally known as the non-insulin dependent diabetes mellitus. Comprehensive elucidation of the genes regulated through this disorder and their wiring will provide a more complete understanding of the overall gene network topology and their role in disease progression and treatment. The proposed strategy was able to find conservative gene modules which play interesting role in T2D, pointing to gene markers such as NR3C1, ADIPOR1 and CDC123. Network-based meta-analysis by enumerating conserved gene modules pave a practical approach to the identification of candidate gene markers across several related transcriptomic studies. The NEMESIS R pipeline for network-based meta-analysis is also provided.