Context: The study and analysis of (meta-)genomics have been providing scientists with valuable insights into the functioning and composition of microbial communities. Latest advancements in next-gen and high throughput sequencing technologies have resulted in significant growth in the data produced and made available for further research. These advancements can help scientists dive deeper into the analysis of uncultivated microbial populations that may have important roles in their environments. Gap: However, analysis of such data requires multiple preprocessing and computational steps to interpret the microbial and genetic composition of samples. For most researchers, configuring these tools, linking them with advanced binning and annotation tools, and maintaining the provenance of the processing continues to be extremely challenging. Moreover, the most common issue with current practices of metagenomics is the reproducibility of the research due to the complexity of setup and configurations. Aim: Our aim is to get a big-picture understanding of the common practices and approaches for metagenomic analyses and to find out which ones are more often used by researchers and why. Further, to compare some of the existing tools and look into possibilities of developing and/or using a reproducible pipeline and give some general recommendations for it. Methods: For this purpose, three main methods were used. First, a literature survey was performed on metagenomic analysis approaches, methodologies, and tools. Next, researchers and scientists with different educational backgrounds active in this field were interviewed. Lastly, the process of pipeline construction and bottlenecks were evaluated through hands-on experience. Findings: By conducting this research, several common pitfalls and shortcomings of metagenomic analysis practices were identified. Since the expertise of most researchers in this field is lacking a fundamental computer science and programming background, very few would attempt developing a pipeline from scratch. Therefore, if instead, they would opt for using “ready-made” General Purpose Pipelines (GPP), they would also face various difficulties in setting up and configuring them to their needs. Also, it has been observed that many of the existing metagenomic tools are not developed and maintained according to computer science code production standards. Therefore, even the more popular tools can suffer from detrimental bugs that can render them broken and consequently deprecated. However, with the emergence of the new “all-in-one” interface-based online platforms such as Kbase.us that enable simple point-and-click set-up and sharing of workflow, there is hope for entering a new era of reproducible metagenomic analysis.

Artificial Intelligence (AI) is increasingly affecting people’s lives. AI is even employed in fields where human lives depend on the AI’s decisions. However, these algorithms lack transparency, i.e. it is unclear how they determine the outcome. If, for instance, the AI’s purpose is to classify an image, the AI will learn this from examples provided to it (e.g. an image of a cow in a meadow). The algorithm can focus on the wrong part of the image. Instead of focusing on the foreground (cow), it could focus on the background (meadow). This way, by focusing on the background, it could produce a false output (e.g. a horse instead of a cow). To show this, an explanation is needed. For this reason, a variety of methods have been created to explain the reasoning behind these algorithms, called explainability methods. In this paper, six local explainability methods are discussed and compared. These methods were chosen as they were the most prominently used approaches for explainability methods for Convolutional Neural Networks (CNN). By comparing methods with analogous characteristics, this paper is going to show what methods exceed others in terms of performance. Furthermore, their advantages and limitations are being discussed. The comparison shows that Local Interpretable Model-agnostic Explanations, Layer-wise Relevance Propagation and Gradient-weighted Class Activation Mapping perform better than Sensitivity Analysis, Deep Taylor Decomposition and Deconvolutional Network, respectively.