Apple’s SimpleFold: Folding Proteins Just Got Easier

Illustration by Mercedes Luna

By ​Kyra Ezikeuzor

Apple has stepped beyond consumer hardware and software to build a fascinating model, SimpleFold, that lets researchers simulate protein folding faster and easier.

SimpleFold is positioned in a field led by Google’s AlphaFold models. Protein folding models are a type of machine learning model known as deep learning models trained on established protein datasets. What’s so useful about these protein folding models is that they can predict the 3D structure of proteins that are essential to drug discovery or fighting antibiotic resistance. Initially, these structures could only have been obtained through time-consuming scientific research, but the creation of models like SimpleFold now expedite that process, improving research in computational medicine and bioinformatics. In fact, John McGeehan, the former director for the Center for Enzyme Innovation, remarked, “What took us months and years to do, AlphaFold was able to do in a weekend.”

In order to understand why Apple’s SimpleFold model is novel, we have to first learn more about Google’s AlphaFold2, the leading model in this field. The AlphaFold2 model is based on prediction. For every new protein, it takes the amino acid coding and compares it to that of other, similar proteins. This process identifies parts of the sequence that change over time. Those parts that change over time are assumed to be the most likely to interact and be located near each other in the protein’s 3D structure.  

So what makes Apple’s SimpleFold model special in comparison? It’s in the name: it’s its simplicity that makes it so special. SimpleFold uses standard transformer layers in its machine learning model, which serve as a primary component of the transformer neural network architecture (Ellmen ‘25).  The neural network is a machine learning model that imitates how the human brain is structured to recognize patterns in data. Additionally, SimpleFold uses flow matching models. These models utilize artificial intelligence to take noise or a simple distribution of possible outcomes and convert it into a complicated distribution of data (Huang, 2024).

Google’s AlphaFold2 doesn’t. Comparatively, Google’s AlphaFold2 model has a more complex neural network architecture. Additionally, Google’s AlphaFold2 is relatively complex because it as a model is trained directly from a large dataset called the Protein Data Bank, a huge database of all experimentally discovered macromolecules (Embl-Ebi ‘25). However, Apple’s SimpleFold doesn’t learn directly from this data but instead relies on the Apple researchers’ “current understanding of the underlying structure generation process into these models.” SimpleFold is less expensive computationally and produces faster results, with the caveat that these results are not as accurate as AlphaFold2’s results (Mendes ‘25). 

Because SimpleFold uses flow matching models, it is less expensive computationally and it provides results faster than models like AlphaFold. This makes it more accessible to students and researchers who may want to test out 3D protein folding for school projects or better visualization for a research project. 

Ultimately, Reducing the computational cost and convolution surrounding research in predicting protein folding can be incredibly beneficial for the future of drug discovery and bioinformatics.

Interested in trying out Apple’s SimpleFold for yourself? You can download and start running the SimpleFold model at Apple’s open source repository. 


Works Cited

Embl-Ebi. (n.d.). What is AlphaFold? | AlphaFold. https://www.ebi.ac.uk/training/online/courses/alphafold/an-introductory-guide-to-its-strengths-and-limitations/what-is-alphafold/

Jia-Bin Huang. (n.d.). How I understand flow matching. (2024, June 2). [Video]. https://www.youtube.com/watch?v=DDq_pIfHqLs&t=702s

Mendes, M. (2025, September 24). Apple develops a lightweight AI for protein folding prediction - 9to5Mac. 9to5Mac. https://9to5mac.com/2025/09/24/apple-simplefold-protein-folding-prediction-ai/

Wang, Y., Lu, J., Jaitly, N., Susskind, J., & Bautista, M. A. (2025, September 23). SimpleFold: Folding Proteins is Simpler than You Think. arXiv.org. https://arxiv.org/abs/2509.18480