Biocentral: Embedding-based Protein Predictions

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Protein Language Models (pLMs) generate powerful protein embeddings.
• Generating embeddings requires significant computational resources and expertise.
• These limitations hinder accessibility for many researchers.

Purpose of the Study:

• Introduce Biocentral, a free and open service for pLM embeddings.
• Simplify the generation and utilization of protein embeddings.
• Facilitate protein prediction and analysis for biologists.

Main Methods:

• Developed a service with standardized access to multiple pLMs.
• Integrated modules for embedding generation, prediction, and model training.
• Provided a graphical user interface (GUI) and programmatic API.

Main Results:

• Demonstrated Biocentral's utility in a large-scale BFVD virus database analysis.
• Successfully reproduced an existing embedding-based prediction method using the training module.
• Showcased the ease of generating embeddings and predictions.

Conclusions:

• Biocentral lowers barriers to using pLM embeddings in biological research.
• The service supports diverse applications, from prediction to model training.
• Biocentral enhances accessibility and usability of advanced protein analysis tools.