DeepMind has published a detailed technical report on AlphaFold 3. This model can predict the 3D structures and interactions of biological molecules such as proteins, DNA, RNA, and ligands, achieving over a 50% improvement in accuracy compared to specialized tools.

In 2018, AlphaFold 1 made a splash in the protein structure prediction competition. By 2020, AlphaFold 2 had largely solved the problem of protein structure prediction. Now, AlphaFold 3 expands its scope to more complex biological molecular systems.

Researchers used AlphaFold 3 to analyze a complex containing both proteins and DNA. The model not only accurately predicted the structures of individual molecules but also revealed how they bind together. This level of detail is crucial for understanding gene regulatory mechanisms.

Another case involved the interaction between proteins and small drug molecules. The structures generated by AlphaFold 3 closely matched experimental data, providing a new tool for drug design.

DeepMind has partnered with the scientific community to offer limited free access to the AlphaFold server. Researchers can submit molecular sequences and receive predictions in just seconds.


AlphaFold 3 is not without its imperfections. It may not always predict rare molecular types accurately, and the model's details are not fully disclosed. However, its emergence has already transformed the way structural biology research is conducted—from purely experimental approaches to a combination of computation and experimentation.

This progress reminds us of the scenario in 1960 when John Kendrew first deciphered the structure of myoglobin. At the time, it took years of experimental work, but now similar levels of precision can be achieved in just seconds. The pace of technological advancement is sometimes faster than we imagine.