Limitations of AlphaFold 2 Modeling
Read the previous section: Five Ways of Using AlphaFold 2
Despite its amazing accomplishments, AlphaFold 2 is not without limitations. Some of these limitations are discussed below.
1. AlphaFold 2 does not predict all structures confidently. When interpreting the results of AlphaFold 2 predictions, it is important to consider the confidence metrics associated with the predicted structures. Generally, regions with pLDDT scores greater than 90 are very reliable, while those with scores less than 50 should be considered unreliable or indicative of disordered regions.
2. AlphaFold 2 does not do protein dynamics. So an AlphaFold 2-predicted structure should be viewed as representing one possible conformation of the protein among many others.
3. Currently, AlphaFold 2 cannot handle protein complexes. However, some researchers in the community have had some success using tricks such as connecting two protein sequences with artificial linkers.
4. AlphaFold2 does not handle domain placement well in certain situations. For example, it may not make accurate predictions for membrane proteins because it was not trained using these proteins under their natural conditions.
5. There are many other things that AlphaFold 2 is unable to do. For example, it cannot handle cofactors, the binding of ligands such as metals, DNA, RNA, and post-translational modifications (PTMs). It also does not make predictions about destabilizing point mutations.
Although AlphaFold 2 is unable to handle certain tasks on its own, it can still be used in conjunction with other structural bioinformatics tools to address these issues. For example, tools like FoldX and Rosetta can use AlphaFold 2-predicted structures as input to predict the stability of a mutated protein. There have also been reports of AlphaFold 2 being used in combination with other tools to improve ligand binding site and structural motif predictions.
Back to the beginning of series: Why Is AlphaFold 2 So Powerful?
About the Author: Justin T. Li received his B.S in Biophysics from Peking University in 1991, his Ph.D. in Neurobiology from the University of Wisconsin-Madison in 2000, and a M.S. in Computer Science from the University of Houston in 2001. He served as an Assistant Professor at the University of Minnesota Medical School from 2004 to 2009, and as Chief Bioinformatics Officer at LC Sciences in Houston from 2009 to 2013. In June 2013, Justin joined AccuraScience as a Lead Bioinformatician. He has published more than 40 research articles in bioinformatics, computational biology, and related fields. From 2013 to 2022, Justin led a team of bioinformaticians at AccuraScience in the completion of more than 120 research and development projects related to bioinformatics, computational and structural biology, and machine learning development. More information about Justin can be found at https://www.accurascience.com/our_team.html.
Need assistance in your AlphaFold or structural bioinformatics project? We may be able to help. Take a look at the intro to our bioinformatician team, see some of the advantages of using our team's help here, and check out our FAQ page!
