Toward computing attributions for dimensionality reduction techniques.

Summary: We describe the problem of computing local feature attributions for dimensionality reduction methods. We use one such method that is well established within the context of supervised classification-using the gradients of target outputs with respect to the inputs-on the popular dimensionality reduction technique t-SNE, widely used in analyses of biological data. We provide an efficient implementation for the gradient computation for this dimensionality reduction technique. We show that our explanations identify significant features using novel validation methodology; using synthetic datasets and the popular MNIST benchmark dataset. We then demonstrate the practical utility of our algorithm by showing that it can produce explanations that agree with domain knowledge on a SARS-CoV-2 sequence dataset. Throughout, we provide a road map so that similar explanation methods could be applied to other dimensionality reduction techniques to rigorously analyze biological datasets. Availability and implementation: We have created a Python package that can be installed using the following command: pip install interpretable_tsne. All code used can be found at github.com/MattScicluna/interpretable_tsne.

Argus: An open-source and flexible software application for automated quantification of behavior during social interaction in adult zebrafish.

Zebrafish show great potential for behavioral neuroscience. Promising lines of research, however, require the development and validation of software tools that will allow automated and cost-effective behavioral analysis. Building on our previous work with the RealFishTracker (in-house-developed tracking system), we present Argus, a data extraction and analysis tool built in the open-source R language for behavioral researchers without any expertise in R. Argus includes a new, user-friendly, and efficient graphical user interface, instead of a command-line interface, and offers simplicity and flexibility in measuring complex zebrafish behavior through customizable parameters. In this article, we compare Argus with Noldus EthoVision and Noldus The Observer, to validate this new system. All three software applications were originally designed to quantify the behavior of a single subject. We first also performed an analysis of the movement of individual fish and compared the performance of the three software applications. Next we computed and quantified the behavioral variables that characterize dyadic interactions between zebrafish. We found that Argus and EthoVision extract similar absolute values and patterns of changes in these values for several behavioral measures, including speed, freezing, erratic movement, and interindividual distance. In contrast, the manual coding of behavior in The Observer showed weaker correlations with the two tracking methods (EthoVision and Argus). Thus, Argus is a novel, cost-effective, and customizable method for the analysis of adult zebrafish behavior that may be utilized for the behavioral quantification of both single and dyadic interacting subjects, but further sophistication will be needed for the proper identification of complex motor patterns, measures that a human observers can easily detect.