This research-to-practice full paper describes our comparative analysis of two different project-based learning (PBL) practices to identify determinant factors in PBL that can motivate and enhance student learning. Project-based learning, a widely recognized form of experiential learning, employs the “learning by doing” approach to help students build concrete knowledge through hands-on experience. Numerous studies have demonstrated that this knowledge transformation process can significantly enhance students' critical thinking and problem-solving skills, while also boosting their motivation and engagement. However, there is a lack of research comparing different PBL practices to pinpoint the essential design elements that maximize student learning. Our study focuses on half-term long PBL practices in online undergraduate database classes with similar demographics, where except for the differences in project-based learning practices, all others in two sections were the same. We compare company-sponsored projects with student-selected projects to analyze how these different approaches impact student learning outcomes and what could be determinant factors influence student project-based learning. Student-selected project-based learning allows students to choose their own data domain of interest. Company-sponsored project-based learning grants students less freedom in topic selection but could provide students with opportunities to collaborate with company professionists and learn the real needs from industry and have a potential to secure an intern job, though there exists uncertainty and unpredictability in the collaboration. Both project-based learning practices vary in the following five characteristics that influence projects: (1) Centrality, (2) Driving question, (3) Constructive investigations, (4) Autonomy and (5) Realism. Our comparative analysis reflects the inherent variations in complexity and difficulty between the two types of PBL practices. We conducted two surveys to investigate student perceptions of their project experiences. We developed a web dashboard to conduct comparative analysis for a more general purpose. Through quantitative analysis of student performance and engagement time, as well as qualitative analysis of survey responses, we identify statistically significant determinants that influence the effectiveness of PBL. This study lays the groundwork for designing guidelines to facilitate effective project-based learning.

Price discrimination has been empirically exposed where e-commercial platforms aim to gain additional profits by charging customers with different prices for the same product/service. This situation becomes even worse in nowadays’ Big Data era, giving the chance for service providers to leverage artificial intelligence technologies to have the deep analysis of personalized patterns, urgently calling for solutions to prevent such discriminated behaviors to protect customers’ rights. This article aims to defend against price discrimination by developing a secure and privacy-preserving solution, provable for e-commerce fairness. Using a newly designed cryptographic accumulator and public bulletin board, our system, called FairECom, allows an auditor (i.e., a customer or third-party auditor) to verify if customers are experiencing price discrimination. In particular, FairECom enables a customer to check if his payment to a product/service is identical to other customers through a privacy-preserving challenge-response protocol, for implementing the price transparency against discrimination. We implement a prototype using an Ethereum-based public bulletin board to conduct the system evaluation. Our evaluation indicates that FairECom can integrate with existing APIs provided by Ethereum and incur acceptable costs when deploying to the e-commercial systems.

In a flipped and adaptive learning environment, adaptability in time is the key to handling constant change and student individuality for their success. To maximize the learning experience, an urgently demanding task is to identify lower-performance students on the fly for targeted timely interventions. Existing state-of-the-art works focus on a regular course setting, where timelines in adaptability is not necessarily a high priority. These works apply batch-based (offline) learning algorithms or ensemble methods, which need the entire training data collected before prediction. These methods, without the whole learning process into consideration, may affect the accuracy of prediction results. In response to the urgent need, we propose to apply an online predictive learning method to handle incoming student data throughout the time steps of a course semester and predict low-performing students for each time step, with a goal to minimize the overall classification error. We built up our experimental design on the multiple learning theories, designed and executed four surveys, and conducted predictive analysis on student data. In the process of feature engineering, we conducted a series of correlation and cause-effect regression analyses and further quantified the determinant factors of predicting student performance. We further developed a framework for identifying low-performing students on the fly and comparing and analyzing deep online learning and diverse traditional batch-based (offline) predictive modeling methods. Our comparative analysis indicates that the online predictive learning approach is encouraging. It outperforms all batch-based (offline) methods overall; prediction results on low-performing students at a time step help identify their problem patterns situated in the context of the whole course progress to design and conduct timely interventions. The innovative study set up a stage for us to deeply understand the learning process, identify main determinant factors, interpret predictive models, and design targeted timely interventions or interactions to help those struggling students through a semester.

In a one-size-fits-all conventional teaching context, student disparities in the background of pre-knowledge, skills, or comprehension lead to the alienation of the struggling students while boring those who are experienced. At issue is the appropriateness of the one-size-fits-all pedagogical model. An alternative is adaptive learning. Dated back to 1912, Edward L. Thorndike proposed the idea of a mechanical miracle that intends to conduct personal instructions through special print. This idea has inspired century-old efforts to automate education by creating such a kind of teaching machine that adaptively features automation, timely feedback, and self-pacing to student mastery in classrooms. With the technology evolving, a great range of state-of-the-art works that provide digital "teaching machines" have emerged, allowing adaptive learning. However, most of them focus on math or elementary reading and writing skills, and few are on programming-based courses. The few available are costly, heavy-weighted in adaptability, or limited in mapping competencies to an entire formal assessment, with questions that are not naturally differentiated in an assessment and thus, lacking flexibility.

Disease development is viewed as a highly dynamic process, where tipping points happen in many complex diseases with irreversible sudden deteriorations in patients’ medical situations. Developing effective strategies for identifying temporal biomarkers, such as identifying temporal protein interaction networks and detecting tipping points, is significantly meaningful for designing an appropriate intervention to prevent the deteriorations. Existing studies inferred dynamical networks from gene expression profiles, uniquely based on static protein-protein interactions, and identified tipping points by a single outlier analysis of dynamic networks under certain assumptions.In this paper, we proposed to identify tipping points of dynamical protein-protein interaction networks through a thermodynamics-enriched ensemble framework In this framework, we constructed dynamical protein-protein interaction networks (DPPINs) by integrating tissue-based direct protein-protein interactions with multiple temporal or non-temporal sources. Through a thermodynamic analysis of these dynamic networks, we identified an abrupt change of free energy at a tipping point and validated this result by relative entropy of dynamic networks.

This Innovative Practice Full paper presents an effort on fostering student motivation and autonomy in support of adaptive learning. Higher education has been transformed by digital technology to be more interactive, self-paced, and adaptive to students. But this technology presupposes that students possess autonomy and are innately well motivated. Unfortunately, many students entering college lack motivation and self-control. These dispositions retard self-paced adaptive learning and limit the effectiveness of imparting improved adaptability in this technology. Taking inspiration from the self-determination theory, we investigated, in the context of adaptive learning, the importance of nurturing student autonomy and enhancing student situated motivation. In this paper, we present our experiential study in a gateway core course of computer science, in which adaptive preparation and learning pedagogy has been adopted for four semesters, one semester without motivating support and others with this support. By comparing and analyzing student engagement and performance over these four semesters, we observe that nurturing student autonomy and enhancing motivation are critical factors in maximizing the effectiveness of adaptive learning.

This Innovative Practice Full paper presents our modeling methods to identify at-risk students early to scaffold for better intervention or interactions with students. We executed the data-driven practice in undergraduate and graduate Data Structures and Algorithms (DSA) courses for multiple semesters. Abstract nature in both courses can often be hard to teach and difficult for students to grasp. To resolve the difficulty and assess its efficacy, we propose predictive modeling methods to identify at-risk students, find student behavior patterns, and distinguish and target the areas that they need, to facilitate scaffolding instructions. We employed two feature selections and four predictive modeling methods, such as support vector machine (SVM), k-nearest neighbors (KNN), naive Bayes classifier (NBC), and random forest. Our comparative analysis on these predictive models indicates that random forest performed best in minimizing false negative (i.e., type II) error while not increasing false positive(i.e., type I) error significantly.

This Research to Practice Work in Progress paper develops an adaptive learning module aimed at enriching data structures and algorithms courses (CS2/DS) with introductory parallel computing. We emphasize exploitable shared-memory parallelism with the intention of teaching students to decompose a problem or its underlying big data structure into parts that can execute effectively in many- or multi-core processes. The module can set up conditional mastery paths and differentiate assignments for individual students automatically. The adaptive learning reflects the student-directed learning by doing and so fits naturally into the conventional CS2/DS courses. We conducted a 50-minute learning session in an adaptive learning CS2/DS class for a group of 56 students in the Fall 2019 semester. The experimental results, from two surveys executed before and after the session, was successful in engaging students, instigating students' interest in parallel computing, and improving the students' awareness and appreciation. The module can be extended to support other parallel computing concepts, such as critical section, race condition, and multi-thread synchronization and cooperation techniques. As a side benefit, we would like to show that adaptive learning prepares students, gets students to engage, and enhances their performance.

In this era of technology and science, data skills are critical for full participation in the workforce and contemporary society. Alarmingly not all graduates exit college with what are nothing less than survival skills. The goals of the course curriculum innovation are to raise the awareness of data science, promote retention, increase the interest and curiosity, and boost essential data skills for all students on campus. These goals are similar to existing efforts on the design of introductory data science courses for majors and non-majors. The distinctiveness of this course curriculum resides in hands-on learning by examples, case studies, and team-based projects within a low-stakes format, where students from different disciplines early in their college careers collaborated to ethically solve problems in the repetition of data science life cycles. The Innovative Practice Category Work in Progress paper presents our experience in integrating introductory data science into a college-wide computer and information literacy course via collaborative practices of exemplar and project-based learning. The collaborative learning with wide interdisciplinary focus enriched our pedagogues and scalability. Its effectiveness has been measured in a poster session and student perceptions on 5-point Likert scales. Furthermore, integrating introductory data science into a computer and information literacy course served to optimize existing educational resources and facilitate multiple pathways in college. The methodology and studio-like training shown in the experience report can be expanded to upper-level data science courses.

In the Big Data era, essential data skills are critical for everybody in full participation of the workforce and modern society to better understand decisions from diverse levels of society. However not all graduates exit college with these needed skills. In this poster, the author(s) report on the integrating of data literacy into a college-wide computer and information literacy course with collaborative practices of project-based learning, which aims at increasing students' awareness of data science and boosting essential data skills of college-wide students. In the poster, we will demonstrate our module design on training essential data skills through examples, case students, and team-based projects within a college-wide computer and information literacy course. We will also present our comparative analysis on one data-enhanced course session with other two course regular sessions as control. In the data-enhanced session, students formed a team, choose a topic in the engaging context, specified questions that they want to ask and answered through data, analyzed data from different perspectives, evaluated results, and present in the course poster session. Two measures are used in assessing the effectiveness of exploratory and collaborative learning in data literacy: student performance in a poster session and student perceptions on Likert scales. Thus far the evaluation results and statistical analysis showed the innovative practice significantly intrigued students' interest in data science and boosted students' fundamental data skills, for both computer science or non-computer science majors.

In a drug development process, appropriate drug-binding selectivity is critical for a success drug. However the selectivity in a data source, showing the intensity of efforts, may be limited to prior knowledge of the expertise or be biased towards the hypothesis testing. With the increasing of drug screening data, it is challenging to coordinate the efforts and execute data governance at a large scale. Visual selectivity analysis for examining target selection is in demand. We proposed a knowledge-driven approach and designed an ontology reference model to provide an intuitive view of the selectivity in a drug-target interaction network data. We employed the model to carry out the visual selectivity analysis on the NIMH Psychoactive Drug Screening Program (PDSP) data and the LINCS Compounds-interacting ChEMBL Database. The analysis indicates the possible `dark matter' drug targets. The approach can be expanded to coordinate other experimental screening data and set a stage for the analysis of the mechanism of action of biological therapies.

With the increasing number of ontologies being designed to represent and manage knowledge in all sorts of sectors, ontology alignment and integration become more and more important in aggregating intelligent efforts on homogenous and heterogeneous data. From the computational perspective, it is challenging due to the ubiquitous existence of diverse classifications of same data. In this paper, we propose an active ontology integration and alignment system, which plugs in expandable learning reference context pool. In the reference context pool, we have integrated WordNet, MeSH, and external curated mapping sources (ICD9 to SNOMEDCT) with an extension to injecting UMLS. The active ontology integration and alignment system takes account of not only subsumption tree but also directed acyclic graph underlying ontologies. It allows 1) finding exact one-to-one matching terms of pairwise ontologies, 2) finding inexact one-to-one term mappings, where two terms have at least a concept in common on basis of the lexical context, and 3) finding one-to-many concept mappings, where one concept can be lexically mapped to the combination of multiple exclusive concepts.

An important issue in protein-protein interaction network studies is the identification of interaction dynamics. Two factors contribute to the dynamics. One, not all proteins may be expressed in a given cell, and two, competition may exist among multiple proteins for a particular protein domain. Taking into account these two factors, we propose a novel approach to predict protein-protein interaction network dynamics by learning from conflict-sensitive phosphorylation dynamics. We built a training model from conflict-sensitive phosphorylation dynamics [3]. In this model, each node is not an individual protein but a protein-protein pair and is labeled with terms representing conditions in which the interaction should be observed. We mapped the protein pairs in a vector space, built hyper-edges over the interaction nodes, and developed rank-like SVM with Laplacian regularizers for PPI network dynamics prediction. We also employed the standard Fl measure for evaluating the effectiveness of classification results.

Some emerging applications deal with graph data and relie on graph matching and mining. The service-oriented graph matching and mining tool has been required. In this demo we present the web service tool WS-GraphMatching which supports the efficient and visualized matching of polytrees, series-parallel graphs, and arbitrary graphs with bounded feedback vertex set. Its embedded matching algorithms take in account the similarity of vertex-to-vertex and graph structures, allowing path contraction, vertex deletion, and vertex insertions. It provides one-to-one matching queries as well as queries in batch modes including one-to-many matching mode and many-to-many matching mode. It can be used for predicting unknown structured information, comparing and finding conserved patterns, and resolving ambiguous identification of vertices.

SUMMARY: The accumulation of high-throughput genomic, proteomic and metabolical data allows for increasingly accurate modeling and reconstruction of metabolic networks. Alignment of the reconstructed networks can help to catch model inconsistencies and infer missing elements. In this note, we present the web service tool MetNetAligner which aligns metabolic networks, taking in account the similarity of network topology and the enzymes' functions. It can be used for predicting unknown pathways, comparing and finding conserved patterns and resolving ambiguous identification of enzymes. The tool supports several alignment options including allowing or forbidding enzyme deletion and insertion. It is based on a novel scoring scheme which measures enzyme-to-enzyme functional similarity and a fast algorithm which efficiently finds optimal mappings from a directed graph with restricted cyclic structure to an arbitrary directed graph. AVAILABILITY: MetNetAligner is available as web-server at: http://alla.cs.gsu.edu:8080/MinePW/pages/gmapping/GMMain.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Network alignments are extensively used for comparing, exploring, and predicting biological networks. Existing alignment tools are mostly based on isomorphic and homeomorphic embedding and require solving a problem that is NP-complete even when searching a match for a tree in acyclic networks. On the other hand, if the mapping of different nodes from the query network (pattern) into the same node from the text network is allowed, then trees can be optimally mapped into arbitrary networks in polynomial time.In this paper we present the first polynomial-time algorithm for finding the best matching pair consisting of a subtree in a given tree pattern and a subgraph in a given text (represented by an arbitrary network) when both insertions and deletions of degree-2 vertices are allowed on any path. Our dynamic programming algorithm is an order of magnitude faster than the previous network alignment algorithm when deletions are forbidden. The algorithm has been also generalized to pattern networks with cycles: with a modest increase in runtime it can handle patterns with the limited vertex feedback set.We have applied our algorithm to matching metabolic pathways of four organisms (E. coli, S. cerevisiae, B. subtilis and T. thermophilus species) and found a reasonably large set of statistically significant alignments. We show advantages of allowing pattern vertex deletions and give an example validating biological relevance of the pathway alignment.

Network mapping is a convenient tool for comparing and exploring biological networks; it can be used for predicting unknown pathways, fast and meaningful searching of databases, and potentially establishing evolutionary relations. Unfortunately, existing tools for mapping paths into general networks (PathBlast) or trees into tree networks allowing gaps (MetaPathwayHunter) cannot handle large query pathways or complex networks. In this paper we consider homomorphisms, i.e., mappings allowing to map different enzymes from the query pathway into the same enzyme from the networks. Homomorphisms are more general than homeomorphism (allowing gaps) and easier to handle algorithmically. Our dynamic programming algorithm efficiently finds the minimum cost homomorphism from a multisource tree to directed acyclic graphs as well as general networks. We have performed pairwise mapping of all pathways for four organisms (E. coli, S. cerevisiae, B. subtilis and T. thermophilus species) and found a reasonably large set of statistically significant pathway similarities. Further analysis of our mappings identifies conserved pathways across examined species and indicates potential pathway holes in existing pathway descriptions.

Parallelization of sequential programs is often daunting because of the substantial development cost involved. Previous solutions have not always been successful, partly because many try to address all types of applications. We propose a platform for parallelization of a class of applications that have similar computational structure, namely graph-structured iterative applications. iC2mpi is a unique proof-of-concept prototype platform that provides relatively easy parallelization of existing sequential programs and facilitates experimentation with static partitioning and dynamic load balancing schemes. We demonstrate with various generic application graph topologies that our platform can produce good performance with very little effort. The iC2mpi platform has a good potential for further performance improvements and for extensions to related classes of application domains.