Learned Embeddings from Deep Learning to Visualize and Predict Protein Sets.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning in Biology

Background:

• Machine learning (ML) and artificial intelligence (AI) models are increasingly used in molecular biology and medicine.
• Language Models (LMs) adapted from Natural Language Processing (NLP) can encode information within protein sequences.

Purpose of the Study:

• To demonstrate the potential of protein Language Models (LMs) for generating descriptive protein representations (embeddings).
• To present simple and reproducible workflows using the bio_embeddings pipeline for generating embeddings and visualizations.
• To highlight the application of embeddings in predicting protein function and structure and in sequence analysis.

Main Methods:

• Utilized the bio_embeddings pipeline and modules for generating protein embeddings from sequences.
• Developed workflows for creating visualizations of protein sequences and their annotations.
• Applied machine learning libraries to use embeddings as input features for predictive models.

Main Results:

• Protein LMs generate embeddings from sequences efficiently, with comparable or improved predictive ability over traditional methods.
• The bio_embeddings pipeline facilitates reproducible generation of embeddings and visualizations.
• Embeddings can be used for predicting protein function/structure, homology inference, and sequence alignment.

Conclusions:

• Protein LMs offer a powerful and efficient approach to understanding protein sequences.
• The bio_embeddings pipeline provides accessible tools for researchers to leverage protein embeddings.
• Further research can harness these tools for novel applications in molecular biology and medicine.