Protein aggregation is a century-old problem that is still poorly understood. It is a major hindrance in the development of protein-based therapeutics and involved in several diseases such as neurodegenerative disorders (Alzheimer’s disease, Parkinson’s disease etc.), type 2 diabetes, AL amyloidosis etc. However, on a positive note, understanding protein aggregation has potential application for the development of commercial amyloid based nano-bio materials. In this work, we have developed computational resources to predict aggregation capability of proteins and mainly focused on understanding the aggregation kinetics of amyloidogenic proteins.
We developed a comprehensive database, “CPAD 2.0”, which contains experimental data on protein aggregation [1]. The database focuses on mechanistic (aggregating proteins and aggregation-prone regions), kinetics (time-dependent intensities and apparent rate of aggregation from fluorescence-based assays) and structures related to amyloids. The database can be accessed at https://urlprotection-tko.global.sonicwall.com/click?PV=1&MSGID=202012230435320171245&URLID=5&ESV=10.0.6.3447&IV=861D2C077669E7FF665D6C5061030AEB&TT=1608698133708&ESN=xtyrjNFgpZWrnpr%2F2gLpVzZk7iKa%2FYY4OBVtikLGvPo%3D&KV=1536961729279&ENCODED_URL=https%3A%2F%2Fweb.iitm.ac.in%2Fbioinfo2%2Fcpad2&HK=0A76802EB913BCAF3376C0E95F1E669A798EE239BD26698413514850C57E2264.
The database was used to develop a sequence-based method to predict the absolute aggregation rate of proteins/peptides using machine learning. The method achieved a correlation of 0.72 for leave-one-out cross-validation using features derived from aggregation propensity predictions, disorderness, experimental conditions and other single amino acid-based features. We have also developed two webservers (i) AggreRATE-Disc, to distinguish between aggregation rate enhancer and mitigator point mutations using sequence information with an average prediction accuracy of 82% for leave-one-out cross-validation [2], and (ii) AggreRATE-Pred, to predict the quantitative change in aggregation rate upon point mutation using structure information with an average correlation of 0.73 for leave-one-out cross-validation [3]. These methods to predict the aggregation rate upon point mutation are dependent on the local conformation at the mutation site and features are unique for each structural class. We further developed a method VLAmY-Pred to predict amyloidogenic light chain sequences in antibodies using conventional aggregation-related features, including hydrophobicity, charge and disorderness. The method showed a prediction accuracy of 80% using leave-one-out cross-validation.
The above-mentioned computational resources are compiled into a pipeline “Amylo-Pipe”, which provides complete information related to the aggregation capability of amyloidogenic proteins. The Amylo-Pipe allows users to predict the aggregation-prone region(s), apparent rate of aggregation and change in aggregation rate upon point mutation. The pipeline is user-friendly, automated and optimized for fast prediction. The webserver for the pipeline is available at https://urlprotection-tko.global.sonicwall.com/click?PV=1&MSGID=202012230435320171245&URLID=3&ESV=10.0.6.3447&IV=3001DBC9BB930DBB0D273B07C11A4357&TT=1608698133708&ESN=0UJlZhiWfGzdNie2LnU8MshnwEA9Nx6xW5Y3ft3Dkt4%3D&KV=1536961729279&ENCODED_URL=https%3A%2F%2Fweb.iitm.ac.in%2Fbioinfo2%2Famylo-pipe&HK=48CD3929043A942B964E9081E87506426E3E2FE9220119B6456FD294CAD79A04/.