Vinhthuy Phan

Computer science's increased recognition as a prominent field of study has attracted students with diverse academic backgrounds. This has significantly increased the already high failure rates in introductory courses. To address this challenge, it is essential to identify struggling students early on. Incorporating in-class coding exercises in these courses not only offers additional practice opportunities to students but may also reveal their abilities and help teachers identify those in need of assistance. In this work, we seek to determine the extent to which the practice of using in-class coding exercises enhances the ability to predict student performance, especially early in the semester. Based on data obtained in a CS1 course taught at a mid-size American university, we found that in-class exercises could improve the prediction of students' eventual performance. In particular, we found relatively accurately predictions as early as academic weeks 3 through 5, making it possible to devise early intervention strategies. This work can benefit future studies on the impact of in-class exercises as well as intervention strategies throughout the semester.

Early intervention is critical in increasing student success in Computer Science (CS) courses, which have attracted a diverse student population. In-class exercises, which are often low-stake and quick assignments, are a popular method for active learning and formative assessment. This study explores the potential of using in-class coding exercises for early intervention in programming courses, particularly before midterm exams. We analyzed historical data from a CS1 course to evaluate whether in-class coding exercises can predict midterm exam performance. Our findings reveal that in-class coding exercises are effective predictors of midterm performance and can serve as valuable tools for early intervention. Specifically, exercise scores and time on task are sufficient indicators of student performance. Although in-class exercises are less powerful predictors than traditional metrics, they offer quicker actionable insights. Additionally, predicting students who are struggling is more feasible than forecasting those who will fail or achieve specific letter grades. This research underscores the potential for designing targeted intervention schemes to support students in CS1 and other programming courses, highlighting the importance of timely and data-driven support mechanisms.

Objectives: This study introduces MetaBIDx, a computational method designed to enhance species prediction in metagenomic environments. The method addresses the challenge of accurate species identification in complex microbiomes, which is due to the large number of generated reads and the ever-expanding number of bacterial genomes. Bacterial identification is essential for disease diagnosis and tracing outbreaks associated with microbial infections. Methods: MetaBIDx utilizes a modified Bloom filter for efficient indexing of reference genomes and incorporates a novel strategy for reducing false positives by clustering species based on their genomic coverages by identified reads. The approach was evaluated and compared with several well-established tools across various datasets. Precision, recall, and F1-score were used to quantify the accuracy of species prediction. Results: MetaBIDx demonstrated superior performance compared to other tools, especially in terms of precision and F1-score. The application of clustering based on approximate coverages significantly improved precision in species identification, effectively minimizing false positives. We further demonstrated that other methods can also benefit from our approach to removing false positives by clustering species based on approximate coverages. Conclusion: With a novel approach to reducing false positives and the effective use of a modified Bloom filter to index species, MetaBIDx represents an advancement in metagenomic analysis. The findings suggest that the proposed approach could also benefit other metagenomic tools, indicating its potential for broader application in the field. The study lays the groundwork for future improvements in computational efficiency and the expansion of microbial databases.

Computer science (CS) relies heavily on teaching assistants (TAs) who are often untrained in CS pedagogy. Existing research on CS TA training typically studies American undergraduate TAs at high-resource universities, ignoring the many universities that use graduate TAs, who are often international students, and that don't have the resources to implement the training strategies discussed in the literature. We describe our approach to implement graduate TA training in a high-diversity, low-resource context. We present a needs assessment, design, pilot test, and deployment of our training course, and discuss implications for other similar departments hoping to train their TAs.

Teaching data science can be challenging partly due to a diverse student population and the difficulty of providing a hands-on coding experience on complex topics. To address these challenges, we introduce a software package that facilitates active learning in the form of in-class coding exercises. This approach provides a much-needed hands-on experience in courses with a diverse student population and a highly technical content. Utilizing a popular cloud-based technology, JupyterHub, this approach enables in-class exercises with personalized feedback from the instructor. We report a classroom experience of using the technology for the first time in a graduate-level Machine Learning course, consisting of a mix of Data Science and Computer Science students. We found that, to a great extent, the course instructor could conduct complex in-class exercises within 10-15 minutes of class time. The instructor was able to understand students' abilities and challenges better and provide them with meaningful personalized feedback as well as group feedback. Students felt that the experience provided valuable hands-on practice, helped them figure out coding mistakes, and prepared them better for homework assignments.

Classifying or identifying bacteria in metagenomic samples is an important problem in the analysis of metagenomic data. This task can be computationally expensive since microbial communities usually consist of hundreds to thousands of environmental microbial species. We proposed a new method for representing bacteria in a microbial community using genomic signatures of those bacteria. With respect to the microbial community, the genomic signatures of each bacterium are unique to that bacterium; they do not exist in other bacteria in the community. Further, since the genomic signatures of a bacterium are much smaller than its genome size, the approach allows for a compressed representation of the microbial community. This approach uses a modified Bloom filter to store short k-mers with hash values that are unique to each bacterium. We show that most bacteria in many microbiomes can be represented uniquely using the proposed genomic signatures. This approach paves the way toward new methods for classifying bacteria in metagenomic samples.