RSLpred2: An Integrated Web Server for the Annotation of Rice Proteome Subcellular Localization Using Deep Learning.

Abstract

Rice is one of the most important staple crops, providing food for more than one-half of the world's population worldwide. Identifying the localization of encoded proteins is the key to understanding their functional characteristics and facilitating their purification. The prediction of protein localization experimentally is time-consuming due to the need for meticulous experimentation, validation, and data analysis; computational methods provide a quick and accurate alternative. We propose RSLpred-2.0, an extension of our previously developed and widely used RSLpred-1.0 tool for annotating the rice proteome. RSLpred-2.0 is implemented in four levels to accurately predict protein subcellular localization. The first level differentiates between single and dual localization with accuracy (97.66% in 5-fold training/testing, 98.12% on an independent data) and Matthews correlation coefficient (0.88 training, 0.90 independent). Single localized proteins are classified into ten classes at the second level, with accuracy (98.33% in 5-fold training/testing, 98.46% on an independent data) and Matthews correlation coefficient (0.95 training, 0.95 independent). The third level categorizes dual localized proteins into six classes with accuracy (99.20% in 5-fold training/testing, 96.75% on an independent data) and Matthews correlation coefficient (0.98 training, 0.90 independent). The fourth level classifies membrane proteins predicted in level 1 into single-pass and multi-pass membranes with accuracy (99.83% in 5-fold training/testing, 98.81% on an independent data) and Matthews correlation coefficient (0.99 training, 0.97 independent). The RSLpred2 tool will help the researchers understand many organelle-specific functions, cellular processes, and regulatory mechanisms essential for plant growth, development, and response to environmental stimuli. The web server as well as its standalone version of the software developed from this study is available freely at https://kaabil.net/RSLpred2/ .