Blog

Amino Acid Starter Kit©, AI, and Authentic Modeling: One Teacher's Experience

Written by 3D Molecular Designs | Aug 31, 2026, 7:53:11 PM

In Sarmistha Ghosh’s AP Biology classroom in Charlotte, North Carolina, protein folding is not just a diagram in a textbook. Students are hunched over pipe cleaners, beads, and Amino Acid Starter Kit© models, arguing about where a polar side chain should go—while an AI prediction of the same protein spins on the screen at the front of the room.

For these students, artificial intelligence is not a magic answer engine. It is one more modeling partner alongside physical kits, simulations, and data sources—something to question, compare, and use to refine their own ideas about how protein structure connects to function.

"One of the most exciting insights has been observing how pairing AI with hands-on molecular modeling, like the 3D Molecular Designs Amino Acid Starter Kit©, transforms abstract concepts into meaningful learning experiences," said Ghosh. "The physical models help students visualize protein structure and molecular interactions, while AI encourages them to ask deeper questions, analyze structure-function relationships, and think like scientists. Together, they create a classroom environment where hands-on modeling and AI complement each other to foster authentic scientific inquiry rather than replace it."

Why Bring AI into Protein Folding?

Protein folding is notoriously abstract for high school students. They must hold in their heads that amino acids differ in polarity and charge, that chains coil and fold in three dimensions, and that tiny changes can ripple up to wholeorganism effects.

In her article “Integrating AI in Research & Project-Based Learning in Biology Instruction” in The American Biology Teacher, Ghosh argues that AI tools like AlphaFold now let
students work with realistic structures that used to be locked behind expensive lab techniques. Instead of only hearing that “structure determines function,” students can see how a particular deletion or substitution reshapes a protein and then connect that change to disease symptoms or drug design.

At the same time, she is careful not to let AI eclipse the conceptual work. In her classroom, AI sits alongside handson models, not above them. Students still build, manipulate, and argue about physical structures before they ever ask what AlphaFold “thinks.”

From Beads and Amino Acid Starter Kit© Models to AI Predictions

Ghosh introduces protein folding with a deliberately low-tech start. Students first explore how disulfide bonds, hydrogen bonds, ionic interactions, and polar versus nonpolar side chains shape a protein’s threedimensional form. They use pipe cleaners, beads, and the Amino Acid Starter Kit©, with color and shape cues to represent different amino acid properties.

That concrete work surfaces common misconceptions: students may oversimplify “hydrophobic” as just “doesn’t like water” or struggle to see how multiple weak interactions collectively stabilize a fold. Ghosh provides scaffolds that prompt them to identify and justify interactions, and she reports that by the end of this segment, most students can successfully identify hydrogen bonding sites and reason about why certain regions tuck inward or remain exposed.

Only after this hands-on modeling phase do students turn to AI. They access sequences from UniProt or the NCBI Protein Database, then generate or retrieve AlphaFold predictions for proteins such as human insulin or CFTR. Students compare their physical models and mental rules about folding with the AIgenerated threedimensional structures, checking where their reasoning aligns and where the AI output pushes them to revise their thinking.

This sequence—hands-on modeling, then AI prediction and comparison—keeps student cognition at the center. AI becomes a testable model, not a black box.

A Case Study: CFTR Misfolding and Cystic Fibrosis

One of the anchor experiences in Ghosh’s unit is a case study on cystic fibrosis. Students retrieve the CFTR protein sequence (UniProt P13569) and examine the AlphaFold predicted structure alongside models they have built with beads, pipe cleaners, and the Amino Acid Starter Kit©.

The class focuses on the well-known ΔF508 deletion. Students trace how losing a single amino acid can destabilize folding, trigger endoplasmic reticulum degradation, and ultimately prevent the CFTR protein from functioning as a chloride channel at the cell membrane. In discussion, they connect this disrupted structure to the physiological symptoms of cystic fibrosis—thick mucus, respiratory challenges, and vulnerability to infection.

Here, AI supports a richer narrative rather than replacing explanation. The predicted structure allows students to visualize what “misfolding” might look like, but they must still articulate how that structural change interferes with channel function and how potential drugs might stabilize or rescue the protein.

Ghosh reports that debating questions such as “How could AI help us design a drug for this misfolded protein?” or “What can we trust in this prediction and what still needs experimental confirmation?” push students into higher-order reasoning about both biology and technology.

Data Sources and Visualization as Modeling Partners

To deepen students’ experience of protein structure, Ghosh layers multiple digital tools into the unit. Students move among:

  • UniProt, to obtain protein sequences and view AlphaFold predictions.
  • The Protein Data Bank (PDB), to explore experimentally determined structures and compare them with AI models.
  • Jmol/JSmol, to manipulate three-dimensional molecular structures interactively.
  • Tinkercad, where students build and present their own 3D representations of proteins.

Using these resources, students walk through a “lesson flow” that starts with raw amino acid sequences and ends with physical and digital models they can explain to peers. For example, they examine insulin sequences from humans, Western lowland gorillas, and Bornean orangutans, highlighting conserved cysteine residues that form disulfide bonds and discussing why these positions are so strongly conserved.

By comparing sequences and structures across species, students get repeated practice connecting amino acid properties to fold patterns and ultimately to function. Visual and tactile representations like the Amino Acid Starter Kit and 3D viewers help them bridge scales—from individual residues to wholeprotein shapes and organismlevel traits.

Where AI Fits—and Where it Doesn’t

In the latter part of the unit, Ghosh introduces more explicitly AIdriven tools, including image recognition and AIassisted sequence analysis. Students can, for instance, use tools like Google Lens or computer vision APIs to analyze photos of their models, label features such as disulfide bridges or hydrophobic pockets, and check whether their interpretations align with AIdetected patterns.

They also experiment with AIpowered sequence alignment and prediction tools, such as ColabFold, to align human insulin with homologs in other primates and generate structural models for each species. Students then evaluate differences in predicted folding, hypothesize about evolutionary or functional consequences, and test those ideas against what they know about protein stability and function.

For advanced or AP students, Ghosh outlines an optional extension into CRISPR and AIsupported guide RNA design. Here, students explore how AI can predict ontarget efficiency, offtarget effects, and editing outcomes, and they connect those capabilities back to the same structure–function logic they used for CFTR and insulin. While not every class will have time or readiness for this segment, she frames it as a way to situate AI within current biotechnology and bioethics conversations.

Throughout, she explicitly surfaces limits and risks. Students discuss when AI tools might discourage critical thinking, how overreliance on automated answers could undermine conceptual learning, and why human oversight remains essential in both research and education. Ghosh emphasizes that “drawing the line” means using AI as a tool to enhance curiosity, reasoning, and creativity—not to shortcircuit them.

What Other Biology Teachers Can Try

Ghosh’s work suggests several practical moves any high school or introductory college biology teacher can adapt, even with limited time or tech access:

  • Lead with a driving question. Start with questions like “How does AI know what a protein looks like?” or “How can a single amino acid change cause disease?” to frame modeling and AI work as ways of answering questions students care about.
  • Pair one handson model with one AI tool. Rather than building an entire tech stack at once, choose a concrete modeling experience (such as using the Amino Acid Starter Kit to fold a short sequence) and then bring in a single AI tool (like AlphaFold) to compare predictions and prompt reflection.
  • Make student reasoning visible before AI appears. Have students predict folding patterns, sketch structures, or build models and explain their choices before they see an AI structure. This keeps AI in a supportive role and makes it easier to spot productive misconceptions.
  • Use AI outputs as texts to critique. Invite students to ask where AI predictions might be uncertain, how confidence scores inform trust, and what data is missing. This positions AI as another model to evaluate using evidence, not an authority to accept uncritically.
  • Connect to AP Biology concepts explicitly. Tie activities back to topics like protein structure and enzymes (AP 2.5–2.6) and gene expression and regulation (AP 3.5), as well as science practices around modeling, data analysis, and evaluating technology’s impact on science.

Even small steps—such as adding an AIpredicted structure comparison to an existing protein folding lab—can help students see AI as a tool they can interrogate and use, rather than a distant or opaque technology.

Want To Go Deeper?

This post is adapted from Sarmistha Ghosh’s article, “Integrating AI in Research & Project-Based Learning in Biology Instruction,” published in The American Biology Teacher, 88(3), 159–165 (2026). You can read the full article in the journal: https://doi.org/10.1525/abt.2026.88.3.159.