How is AlphaFold 2 used by scientists?

AlphaFold2 has been used in a wide range of research applications, with impressive results. AlphaFold2 is not intended as a rival to experimental methods. Instead, the starting point provided by AlphaFold2 is facilitating experimental structure determination, assisting with interpretation of low-resolution structural data, and accelerating structural studies. Also, AlphaFold2 can provide testable hypotheses that guide biochemical and cell biology experiments.

Structural biology

Structural biology has been the research field most impacted by AlphaFold2. Predicted protein structures have been used to enhance or fill in experimental structures obtained by techniques like X-ray crystallography, cryo-EM and NMR spectroscopy. Structures from AlphaFold2 have been instrumental for X-ray crystallography studies, serving as a valuable starting point in the process (helping to solve the phase problem via molecular replacement).

AlphaFold2 has enabled researchers to describe the structures of very large protein complexes. One example is the nuclear pore complex, which lines the nucleus of eukaryotic cells and contains around 1000 nucleoporin proteins of about 30 types. Analyses incorporating AlphaFold have now resolved about 90% of the human nuclear pore complex (Mosalaganti et al., 2022). Similarly, AlphaFold2 was instrumental in resolving the structure of Mce1, a protein used by the tuberculosis bacterium to scavenge nutrients from host cells (Chen et al., 2023).

AlphaFold2 predictions have demonstrated great synergy with cross-linking mass-spectrometry (XL-MS). Predicted structures have helped interpret and make sense of distances obtained with this method. XL-MS distances, in turn, provide experimental validation of the predicted models, including predictions for protein-protein complexes. In one study, AlphaFold2 and XL-MS were used to identify the structures and positions of intracellular proteins in the cilia of Tetrahymena thermophila (McCafferty et al., 2023). A similar two-pronged approach combining AlphaFold2 and XL-MS revealed protein-protein interactions in human cells, including “28,910 unique residue pairs captured across 4,084 unique human proteins and 2,110 unique protein-protein interactions” (Bartolec et al., 2023).

Using AlphaFold2 structure predictions to guide molecular and cell biology

AlphaFold2 is proving particularly valuable in guiding mutational analyses. It is also helping scientists generate and test hypotheses about where and how a protein might interact with other proteins or molecules.

When searching for proteins that might perform a desirable function, AlphaFold2 has a role to play in conducting early, low-cost assessments. One such challenge is identifying enzymes that can break down poly(ethylene terephthalate) (PET), a widely-used plastic. Researchers used AlphaFold2 to help them identify 74 potential PET-degrading enzymes and to characterise their 3D structure and mechanisms (Erickson et al., 2022).

AlphaFold2 predicted structures are being used to propose mechanisms of protein action, which are crucial for early-stage biomedical research. For instance, mutations in the human protein PINK1 cause autosomal recessive early onset Parkinson’s disease. Researchers used AlphaFold2 to predict the structure of PINK1. This provided evidence that human PINK1 uses the same mechanism as insect PINK1, which is better characterised both structurally and functionally. The human pathogenic mutations could then be analysed in the context of the protein’s structure (Gan et al., 2021).

It has also proved possible to use AlphaFold2 to predict protein-protein interactions, which are central to many cellular functions. A study in yeast identified 106 protein assemblies that were previously unknown, and a further 806 that had not been structurally characterised (Humphreys et al., 2021). Another study used AlphaFold-Multimer to identify the mechanism of action of the replication factor DONSON. A small in silico screen for DONSON interactors suggested that DONSON catalyses the final step in assembly of the replicative CMG helicase (delivery of the helicase co-factor GINS), a model that was confirmed using structure-function analysis. This study showed that AlphaFold-Multimer can identify new, functionally-relevant protein-protein interactions. (Lim et al., 2023).

Protein engineering and design

AlphaFold2 can be the starting point for protein engineering and design. One group re-engineered a molecular “syringe”, used in nature by bacteria, to deliver therapeutic proteins into human cells. No experimental structural information was available for the syringe “tip”, so AlphaFold2 was used to guide the engineering (Kreitz et al., 2023).

Another group used AlphaFold2 to “hallucinate” symmetric protein assemblies that did not exist in nature. The team then created ten of these artificial protein assemblies and showed that their experimentally-determined structures matched those predicted by AlphaFold2 with an average RMSD of 0.6 Å (Wicky et al., 2022).

AlphaFold2 has also been used to validate engineered variants of proteins. For instance, one group is creating proteins to capture solar energy. They designed a protein maquette to recreate a compact photosynthetic reaction, and used AlphaFold2 to verify its structure (Ennist et al., 2022).

---

For more ways to use AlphaFold2 predicted structures to help tackle fundamental biological questions, see section “Advanced modelling and applications of predicted protein structures“.

Dive deeper into the science and see how AlphaFold is transforming research on the Google DeepMind blog: AlphaFold: a solution to a 50-year-old grand challenge in biology.

Watch the videos in this playlist to discover how scientists are using AlphaFold2 to fuel their research.

Test your knowledge