Back

ESM-PVP: Identification and classification of phage virion proteins with a large pretrained protein language model and an MLP neural network

Li, B.; Liang, G.

2023-12-30 bioinformatics
10.1101/2023.12.29.573676 bioRxiv
Show abstract

Bacteriophages, also known as phages, are essential for the stability of the microbiome system due to their ability to infect prokaryotes, another significant component of the microbiome. Thus, understanding the functions of phage proteins could help us unravel the nature of phages and their roles in the microbiome. However, limited by the low throughput of experimental techniques, a vast number of phage proteins remain unannotated in terms of their functions. Computational methods are expected to solve this restriction due to their high throughput and cost-effectiveness. In this study, we focused on one aspect of functional annotation for phage proteins, the identification and classification of phage virion proteins, and the integration of a large pretrained protein language model and an MLP neural network dramatically improved the performance of these two tasks. Additionally, we compared our model with some previous deep learning models using a newly collected, independent benchmark dataset, demonstrating the strong generalization ability of our model for both tasks. The source codes of ESM-PVP and the software for the PVP identification task have been uploaded to: https://github.com/li-bw18/ESM-PVP.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.