What digital twins and plant genetics reveal about the science skills students actually need

Digital twin modeling is having a moment in biotech. Researchers build a working computational replica of a biological system, then run experiments on the replica in software before touching a lab bench. Consider the guard cell. When a plant senses drought, the hormone ABA sets off a signaling network inside the leaf's guard cells that closes its pores to conserve water. Researchers at Penn State built a Boolean model of that network, a map of which components switch each other on and off, and used it to run experiments in software: knock out a component, predict what happens to stomatal closure, then test the prediction at the bench and revise the model. First published in 2006, it has been refined for nearly two decades as new biology comes in.

It is also, in a simpler form, exactly what students can do in ModelIt!

What a digital twin actually is

The term gets used loosely, so here is a precise version: a digital twin is a computational model that mirrors a real biological or physical system closely enough that you can run experiments on the model and learn something real about how the system behaves. In plant science, that can mean modeling regulatory networks, the relationships between the components that switch each other on and off in response to signals like light, temperature, or water availability.

When a student builds a model of how a plant responds to drought, they are not clicking through a pre-built animation. They are deciding which genes matter, how those genes interact, and what happens when one of those relationships breaks down. That is systems thinking. That is the work.

Why this matters for computational biology careers

The demand side of this is not hypothetical. The U.S. Bureau of Labor Statistics projects that employment of computer and information research scientists, the category that includes computational and bioinformatics roles, will grow about 20% between 2024 and 2034, much faster than the average for all occupations (BLS Occupational Outlook Handbook).

The 2024 Nobel Prize in Chemistry points the same direction. Half of it went to the team behind AlphaFold, a computational model that predicts protein structures in minutes that once took researchers years to resolve in the lab, and the other half recognized computational protein design (NobelPrize.org). The McKinsey Global Institute estimates that generative AI could create $60 to $110 billion a year in value across the pharmaceutical and medical-products industry, much of it by accelerating drug discovery (McKinsey). All of it depends on people who can build and interpret models, not just run code someone else wrote.

The World Economic Forum ranked analytical thinking the most in-demand skill globally in its 2025 Future of Jobs report, with 39% of current workforce skills expected to change by 2030. Meanwhile the entry point that used to be "learn to code" is itself being automated. Google now reports that roughly 75% of its new code is AI-generated and reviewed by engineers, up from 50% in late 2025 and about 25% in early 2024 (Fast Company). What is not being automated is the judgment to know what to model, which relationships to include, and what the output actually means.

What students are building in real classrooms

In ModelIt!, a student working on plant genetics might build a digital model of how a drought-stress signal activates a stomatal-closure pathway. They set the logic: if a given gene is active and water availability is low, another gene switches on. They run the simulation, observe the system behavior, and then ask what happens if they knock out that second gene entirely. It is the same basic workflow a computational biologist uses. No coding, no advanced math. If a student can drag, drop, and connect components, they can build the model.

This is not a simulation students watch. It is a model students construct, which is a meaningful difference. A peer-reviewed study found that undergraduates taught with online computational models and dynamical simulations outperformed peers taught the traditional way (Booth et al., 2021). The principle is not age-specific: students learn a system more deeply by building it than by watching it run.

The NGSS connection teachers actually need

Science and Engineering Practice #2, Developing and Using Models, is one of the most frequently cited and least-delivered standards in K-12 science. Most curricula offer pre-built diagrams students label or simulations students observe. Neither of those is developing a model in the way NGSS intends.

When a student builds a plant-genetics model in ModelIt!, they are identifying variables and the relationships between them, making testable predictions from the model's behavior, revising the model when results do not match expectations, and explaining a complex biological system using a representation they built themselves. That is SEP #2 delivered, not approximated. ModelIt! lessons also include embedded CAST-aligned assessments, which matters for California districts working toward that standard of rigor.

From classroom to research pipeline

ModelIt! is built on Cell Collective, a research-grade computational biology tool used in peer-reviewed systems-biology work. Its defining feature is the one that matters most for students: you build and simulate models without writing equations or code.

Students are not working with a simplified toy version of science. ModelIt! is built on the same underlying modeling approach researchers use, adapted for K-12 without stripping out the rigor. A student who builds a regulatory network in 8th grade is using the same modeling logic a computational biologist applies to drug-target discovery. That continuity is a design decision, not a slogan.

The gap between what students learn in school and what research-grade science actually requires has always been wide. Digital twin modeling is one place where it is visibly narrowing, in classrooms, right now.