Benchmarking AlphaFold2-RosettaRelax for Predicting the Impact of Indel Variants in DHFR

Abstract

In this study, we evaluated a computational method combining AlphaFold2 (AF2) and RosettaRelax (RR) to predict the impact of insertion and deletion (indel) mutations on the stability of dihydrofolate reductase (DHFR). We generated a series of DHFR variants, each containing a single amino acid insertion or deletion, and predicted their structural stability using AF2 to predict the fold and RR to refine the structures and calculate the change in free energy (ΔG). The predictions were compared with experimental data acquired with the CPOP system, to evaluate the performance of the method. The AF2 predictions included the pLDDT score to assess the confidence in the predicted structures, while the RR provided ΔG scores to estimate the stability of the variants. Additionally, we ran RR on DHFR complexed with methotrexate (MTX) to examine how the presence of this ligand affects protein stability. Our results indicate that the AF2-RR combined protocol is effective in predicting the stability changes due to indel mutations, but no significance improvement was found in accuracy when using MTX in the RR predictions. However, the method shows limitations in accurately predicting instability caused by mutations in the core region of the protein, though this might be specific to DHFR. We also observed that indels located within secondary structures such as helices or sheets were generally non-tolerated. This study highlights the utility of AF2 and RR in structural biology and provides insights into the stability of DHFR variants. Furthermore, the availability of the CPOP indel dataset allows for the testing of other computational methods, such as INDELi-X, which combines substitution scores with indel location to enhance prediction accuracy.