Most science students spend years watching other people's models. They click through simulations, observe virtual reactions, and follow pre-set variables to pre-set conclusions. The content is accurate. The thinking is often passive. The research on modeling-based instruction points to something more demanding, and more useful, for students.
Modeling-based instruction asks students to construct, test, and revise their own models of complex systems. A growing evidence base, much of it built on Cell Collective, the research-grade engine underlying ModelIt!, helps clarify what that approach does and does not do. Here is what the studies actually show, including their limits.
A note on scope before we start: the controlled studies below were run in undergraduate biology and biochemistry courses using the same modeling engine that ModelIt! adapts for K-12. The build, test, and interpret workflow is the same. The students in these particular studies are older, which is worth keeping in mind when reading the numbers.
Working with a model engages the brain differently than reading about one
Clark, Helikar, and Dauer (2020) used fMRI imaging to compare two groups of students. One group simulated a computational model of a gene regulatory system. The other read an expert analysis of the same system. When the students later reasoned about that system, the simulation group showed higher activation in several brain regions than the reading group.
One detail deserves to be stated plainly, because it is easy to overclaim: both groups answered the follow-up questions with similar accuracy. The study does not show that modeling produced higher test scores here. What it documents is that working with a model recruits the brain differently than reading does, and that accuracy on model-based reasoning correlated with activity in regions tied to higher-level reasoning. It is evidence about how students engage, not a finished case about outcomes.
The learning gains show up in a controlled comparison
Booth et al. (2020) tested computational modeling modules built in Cell Collective against a section of the same course that received standard instruction (a "no module" control). Students who completed the modules showed statistically significant gains of roughly 7% to 9% across assessments, while the control group's gains were statistically indistinguishable from zero. The difference held after controlling for prior performance and demographic variables.
The study also examined gender, and the finding is more interesting than a flat "no bias." Female students in the no-module course consistently scored lowest on metabolism topics, in line with a documented gap in biochemistry performance. With the modules, and especially with repeated exposure across two courses, the gap between male and female students narrowed. The authors state it carefully: repeated use of the modules may help make learning gains more equitable. That is a real equity signal, reported at the strength the data supports.
(Citation housekeeping: the publicly available version of this study is a 2020 preprint. If a later peer-reviewed version exists, cite that one.)
Where systems thinking fits
Modeling is one way to teach a broader skill: thinking in systems, meaning interconnected components with feedback loops rather than linear cause-and-effect. Systems thinking is increasingly named as a workforce skill in its own right. The World Economic Forum's 2025 Future of Jobs Report lists it among the skills rising in importance through 2030.
This is the gap that a lot of K-12 science instruction leaves open. A student can memorize the steps of the cell cycle and still not see why disrupting one regulatory protein can cascade into cancer. Building and testing a model is one of the few classroom approaches that makes those connections visible, because the student has to specify the components, connect them, and watch what happens when one changes.
What predicts whether teachers keep using a tool
Song et al. (2023) studied the factors behind instructors' adoption of Cell Collective using a technology-acceptance model. The study separated a tool's usefulness and ease of use into two sides: how it serves teaching, and how it serves learning. The learning-side factors, how useful and how easy to use the tool was for student learning, predicted instructors' intention to adopt. The teaching-convenience factors, on their own, did not.
The practical read is not "ease of use doesn't matter." It does, on the learning side. The signal is that instructors weigh what a tool does for students heavily when deciding whether to keep it. That suggests a useful set of questions for evaluating any science platform:
- Does it ask students to make decisions, or just click through?
- Does it produce artifacts you can assess, or just completion data?
- Can students explain what they built and why it behaves the way it does?
- Does it align to specific NGSS practices, or just gesture at them?
NGSS Science and Engineering Practice #2 (Developing and Using Models) and SEP #5 (Using Mathematics and Computational Thinking) are not satisfied by watching a simulation. The standard asks students to develop and use models, which is a construction task, not a consumption task.
Why this matters more now than it did five years ago
In 2024, the Nobel Prize in Chemistry recognized computational protein modeling. Half went to David Baker for computational protein design, and half to Demis Hassabis and John Jumper for AlphaFold, which predicts protein structure (Nobel Prize in Chemistry 2024). The World Economic Forum's 2025 report ranks analytical thinking as the most in-demand skill, with about 7 in 10 employers calling it essential, and estimates that 39% of workers' core skills will be transformed or outdated by 2030. Computational and information-research roles tied to this work are projected to grow about 20% over the decade, much faster than the average for all occupations (U.S. Bureau of Labor Statistics).
The students in your classroom now will enter a workforce where building and interpreting models is a normal scientific skill, not a graduate specialty. The research on modeling-based instruction is not a finished proof, but it points in a consistent direction, and it lines up with where the work is heading.
The question worth asking about any science tool is not whether it covers the standards. It is whether students are doing the thinking that makes the standards worth covering.