Enhancing ProteinBERT: Integrating Intrinsically Disordered Proteins for Comprehensive Proteomic Predictions

While advances have been made in the study of globular proteins, proteomics research has yet to fully address the unique realm of intrinsically disordered proteins (IDPs). In terms of structural and functional dynamics, IDPs differ from their globular counterparts. This article presents two major contributions: (1) the development of extensive IDP protein datasets for enhanced training, and (2) the enhancement of ProteinBERT’s predictive capabilities to also include IDPs along with globular proteins. As a result of the refinement of our model, we have emphasized: 1) homology predictions, which are important in light of protein sequence evolution; 2) the possibility of secondary structures; and 3) the determination of folding classes. By fine-tuning ProteinBERT, prediction accuracy is markedly improved, highlighting the importance of diversified datasets in advancing proteomic machine learning.