The Digital Imaging and Vision Applications in Science program was designed to build the computational self-efficacy and skill of students majoring in biological and chemical sciences within their first two years of higher education. A previous report of a three-year pilot study indicated that the program interventions caused significant gains in both areas. The community of practice supported by the DIVAS program supported parallel initiatives to broaden participation through expanding access to the image processing workshop that forms the core of the program, supporting faculty and secondary educators in developing computing modules for their classrooms, and in creating and staffing a ‘writing center for computing’ on the host campus. Overall, the DIVAS program has supported a local culture of computing. To support similar development elsewhere, the DIVAS program interventions and community elements are freely available for adoption. The growth of the DIVAS community can provide opportunities that would have an additional positive impact on students as well as strengthen the support available for educators and facilitators.

The generation of data in natural, physical, and computer sciences is now so massive that data analysis, interpretation, and visualization often cannot currently keep up with the rate that it can be routinely produced [1]. This data explosion has necessitated interdisciplinary collaboration to build tools, models, and visualizations well-matched to the massive, complex, multi-faceted data that are out there to explore. Through this interdisciplinary work, scientists are developing new tools enabling observation of aspects of our universe that have previously been hidden or beyond our reach. The current potential to expand our understanding is much like it was when the microscope turned science on its head centuries ago [2].

Education and training within the natural and physical sciences are largely insufficient to equip individuals with the necessary experience and skills to meet emerging data processing and analysis needs [3–6]. There is a strong consensus across the natural and physical sciences to improve the computational competency of students so that meaningful interdisciplinary work can occur and students' full potential as innovators can be realized. To these ends, the Digital Imaging and Vision Analysis in Science (DIVAS) project was created.

DIVAS secured funding in 2016 with funding from the National Science Foundation’s Improving Undergraduate Science Education division [7]. The project was designed to engage novice learners (mostly first- and second-year college students) from chiefly biological and chemical sciences in computational thinking and coding using image data as a ‘hook’. Images are widely used for diagnostics, phenotyping, and analytical measurements. They are also easy to obtain and their utility in making relevant observations readily accessible to novice learners. Finally, analysis of image data provides opportunities for learners to engage in all of the aspects of computational thinking including recognition of a problem, analysis of solutions, design of a solution, and implementation and testing.

Our pilot study indicated that a series of scaffolded interventions experienced within the context of a community of practice can be effective in improving the self-efficacy and computational thinking skills of students not normally represented in computer science courses [8]. The pilot study has motivated an effort to see how broadly these results can be replicated. In this report, we will discuss the key elements of the DIVAS project and will discuss how the dissemination of program elements is informing how computational self-efficacy leading to computational thinking skills can be fostered more broadly across the natural and physical sciences.

The DIVAS program is targeted toward first-year students enrolled in introductory biology and chemistry courses. The program name, leadership composition, and recruiting strategies resulted in a high percentage of individuals who identify as women joining the program (76% of all program scholars) (PLoS). The program begins in the spring with a one-credit seminar and continues 5-8 weeks into the following summer. The program ends with a capstone seminar the following fall or spring. From the beginning, students join a community of practice that will support them throughout the rest of the program in the first DIVAS seminar. Scholars also learn about images as a form of data and do some basic coding, meet a diverse array of professionals who use coding in their work and explore the working environments where coding occurs.

Summer programming begins with an intensive, one-week coding workshop that is also open to community members broadly. The workshop is composed of two days on basic Python operations, code reposition using git, and bash shell functions, then three days on image processing using Scikit-image libraries. Participants are situated within the room to encourage collaboration with neighbors and are presented with a problem at the beginning of each major portion of the workshop that can be addressed with the functions and tools they learn. By the end of the workshop, participants are able to write scripts to count the number of colonies on a plate and to track the progression of a titration over time using a handful of basic image processing functions.

Following the coding workshop, scholars enter into four weeks of pair programming divided into a pair of two-week sessions. Scholars are given a common challenge at the beginning of each session that they discuss and ask questions about. From that discussion, major goals for what the group would like to achieve by the end of the two weeks are determined. Scholars learn about pair programming and are divided into pairs to devise and implement their own strategy for solving the problem. Each day the group meets for a stand-up meeting to report on progress and daily goals. At the end of each week, code is pushed to a common repository and teams print out and annotate each other’s code for review. The code reviews help pairs set their goals for the following week or wrap up their code depending on which week of the project they are in. In the second two-week session pairs are reassigned and the same routine is repeated.

The program ends with another one-credit seminar. In this seminar, scholars revisit the DIVAS repository to clean up and annotate code, review the coding workshops and make adjustments based on their experience and suggestions, and they meet with incoming scholars to welcome them within the community and provide support as needed. Scholars also learn about parallelization and gain experience with parallel programming, which is relatively straightforward to apply with the image processing scripts they’ve already written.

The DIVAS pipeline was tested on three cohorts of up to 6 scholars over three years. A total of 17 scholars participated, 14 of which were from the home institution, Doane University, a private liberal arts college in Nebraska. The other three scholars came from our partner campus at St. Edward’s University, also a private liberal arts college as well as a minority-serving institution, in Austin, TX. Most scholars identified as women (76%) and were in their first year of college (82%) majoring in biological or chemical fields (82%). Detailed results of this study have previously been reported [8]. Overall, we saw self-efficacy in computing increase by 34% on average as well as significant growth in all areas of computational thinking measured (recognize the problem, analyze solutions, design a solution, implement a solution) after the coding workshop [8]. It is interesting that self-efficacy grew even while interest in pursuing careers using computational skills did not and coding tasks became more challenging. Details of the pilot study, including resources and information about each of the DIVAS program interventions can be found in our previous publication [8].

Image processing has become routine in studies that aim to understand atomic, molecular, and cellular dynamics, associate genomic elements with phenotypes of interest, in breeding programs, and in a variety of monitoring and modeling in fields such as agriculture, ecology, and drug development. This increased demand within the scientific community for image processing skills led to the adoption of the image processing elements of our workshop into a Data Carpentry lesson [9,10]. Data Carpentry supports community-driven development of domain-specific lessons to support research training needs.

The lesson is still in the early adoption process, which began by converting it from using OpenCV libraries to Scikit image libraries, which are much easier to implement across a range of platforms and environments. It has been tested at three research institutions in the United States and Germany. The lesson assumes basic knowledge of Python, git, and bash and covers the basics of image processing including image representation, creating histograms, blurring and thresholding, drawing and masking, edge detection, and object segmentation using connected components analysis. The two challenges that DIVAS scholars work on in this portion of the workshop are also available in this lesson. The Data Carpentry lesson is currently maintained by DIVAS project investigator Mark Meysenburg, but as it is more widely utilized by others in the community and broader community needs are identified, its composition and approach will ideally be modified to meet those needs.

The Computing Center for the Liberal Arts. An important consequence of the implementation of the pilot study has been the creation of a broader community of students with computing skills on the Doane University campus. No longer siloed into specific departments and programs, students who may not have normally interacted with each other academically were now connected through common interests and skills through their curricular and co-curricular work.

Recognizing the ways DIVAS scholars could or were broadening their community of practice to include peers who needed to build their own computational skill as well as peers with more expert knowledge that could provide support, the DIVAS team created a ‘writing center for computing’ at Doane called the Computing Center for the Liberal Arts (CCLA). The CCLA is a place for anyone within the Doane community to get feedback and assistance on any computing project from setting up an Excel spreadsheet to using Doane’s supercomputer Onyx for a research need.

Support from the National Science Foundation enabled the team to hire a center director whose duties included packaging program structural elements for broad implementation in a ‘computing center in a box’ [11]. Like writing centers across the nation, the CCLA is fueled by peer-to-peer support. Peer instructors develop short tutorials and guides on common computing needs, answer questions and provide support via a Slack workspace and email address, and hold a physical space in the learning commons to provide face-to-face support. The CCLA is now in its third year. At least one in four DIVAS scholars have served as peer consultants and one former scholar is the current training manager for eleven new consultants. Individuals majoring in six different disciplines have utilized its services. The center is growing steadily as the campus community becomes more familiar with its goal. Special efforts have been made to reach out to the humanities and social sciences to support their computing needs as well.

Introductory biology and chemistry courses. Computational thinking through image analysis has been brought into introductory biology and chemistry courses in two ways. First, an image processing module was designed for general chemistry to investigate the hydrophobicity of materials by measuring the contact angle of a drop of water upon them. Students take pictures of the water droplets they added using their smartphones and then ImageJ to measure the angle of incidence from each image. This module has now been used for over five years.

In an inquiry-based introductory biology course, mostly first-year students used a Google Colab notebook written by former DIVAS scholars to analyze images from a system to measure bacterial movement toward molecules (chemotaxis). A drop of agarose containing a test molecule or saline (control) is added to the center of each well in a six-well plate. The solid agarose droplet is then surrounded by stained E. coli in saline. Images are taken over time using flatbed scanners. Students first use ImageJ to develop their own strategies for measuring the change in cloudiness around the agarose plug over time (an indication that chemotaxis has occurred). As completely naive coders, they then reviewed the code in the Colab notebook by drawing out a flowchart of what they see the code doing. When students were unsure of how the code was functioning, they tried changing part of it to see what effect it had. As a group, the class was able to add clarifying annotations and had created a full code map. From there, students used this code and modified parameters as needed to measure the change in cloudiness around the agarose plugs in their own well plates.

Teacher training workshop. Finally, secondary educators in the state of Nebraska had the opportunity to use another Google Colab notebook written by former DIVAS scholars last summer. Scholars wrote this code to measure the height and density of invasive grass growing in pots. The grass system is flexible in providing a variety of angles from which to engage students in inquiry-based learning and to learn the concepts of asking questions, designing experiments, and analyzing and representing data. Teachers explored the pros and cons of manual measurements, then reviewed the image-based approach to taking similar measurements by creating their own code maps as they worked through the Colab notebook. As teachers worked through the code they identified other approaches toward taking the same measurements and ways to test the code at different points to make sure it was functioning. This group of teachers, the large majority of whom had never coded, were able to use the notebook and to evaluate it for how it might be used in the classroom both as a measurement tool and as a source of inquiry-based curriculum within a matter of a couple of hours.

Over the three years of the pilot study, we have learned a lot about how self-efficacy in computing could be fostered early in a student’s college career. The pilot data and anecdotal experiences themselves might inspire other schools to try similar approaches. However, there is also a need to more fully test the DIVAS program interventions across institutional types and different student populations to better understand the generalizability of what we have observed. Broader implementation of all or part of DIVAS program elements will also give us a better understanding of the most impactful elements of the program so that those schools wanting to implement a more streamlined program can still realize significant impacts on their students. A broader DIVAS alliance also provides additional opportunities for students to form collaborations that can further build their own skills while also strengthening the community of practice. To additionally broaden the DIVAS community of practice, we also hope to expand the teacher training workshops, including both secondary and undergraduate educators, in the Training of Trainers (ToT) model so that educators are empowered to integrate DIVAS interventions into their classrooms and research labs.

There is a long-standing need to find ways to broadly infuse computational thinking, coding, and the use of scientific software within natural and physical science undergraduate education. Our experience in implementing the DIVAS project, our pilot study, and the additional opportunities it fostered suggest that this can be done in environments where a community of practice approach can be supported. We have begun to see how that community can change the way students view computing from a specialized skill to a set of tools anyone can learn to use. We have seen that novice learners can learn to effectively use computational tools to solve problems relevant to their disciplines, thereby gaining confidence in their ability to conduct additional computational work and highly desirable workforce skills. The generalizability of our experience and the essential elements leading to improved computational self-efficacy and skill among novice learners will become more clear with the broader adoption of DIVAS program elements.