Practical Applications of Language Models in Protein Sorting Prediction: SignalP 6.0, DeepLoc 2.1, and DeepLocPro 1.0

Protein sorting prediction, or protein subcellular location prediction, is an important task in bioinformatics. It is also a task where protein language models have proven their value in practice. SignalP 6.0 (for prediction of signal peptides), DeepLoc 2.1 (for prediction of subcellular location and membrane association in eukaryotes), and DeepLocPro 1.0 (for prediction of subcellular location in prokaryotes) are all based on large pretrained protein language models. This has led to substantial improvements over previous methods, especially for rare classes where available labeled data are limited.