Understanding the mechanisms of protein folding has been an everlasting endeavor of the scientific world. That said, accurately predicting protein structures from their amino acid sequences by computational means is a milestone drawing closer. Achieving that would shed light on countless molecular mechanisms related to protein function and revolutionize modern medicine.
Following the emerging advancements in Machine Learning, participants of the Critical Assessment of protein Structure Prediction (CASP) have been developing increasingly better algorithms for protein structure prediction. AlphaFold is an outstanding example of such a method that achieved near experimental accuracy in CASP14. Since its release, AlphaFold has been widely used among the scientific community for the prediction of individual structures, complexes, or even whole genomes. However, despite its phenomenal performance, limitations still apply to it and so, validation of the predicted structures is deemed necessary.
In this postgraduate dissertation study, ColabFold (an implementation of AlphaFold which utilizes MMseqs2) is employed for the prediction of structures of a set of proteins related to the pathology of Alzheimer’s disease. These structures were then superposed to a set of experimental structures from PDB and comparison metrics were computed to evaluate prediction accuracy. To increase accuracy, AlphaFold was configured to execute a higher number of recycles than normally programmed. Finally, examination of the structures was performed to identify existing position-specific deviations among predicted and experimental models. The results suggest an overall high level of accuracy in AlphaFold’s predictions of the selected AD related proteins with very little deviation from existing experimental models.
In addition to all the above, the present work also serves as a review of a selection of state-of-the-art protein fold recognition methods developed over the past five years.