UPCLASS: a deep learning-based classifier for UniProtKB entry publications.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Publications in the UniProt Knowledgebase (UniProtKB) are categorized by evidence type (e.g., function, interaction, expression).
• Systematic categorization of computationally mapped bibliographies is needed for efficient curation.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) model for classifying publications linked to UniProtKB accession annotations.
• To address the challenge of publications being associated with multiple protein accessions and evidence categories.

Main Methods:

• A CNN model was designed to divide documents into sections relevant to protein annotation evidence.
• Feature sets were created for each accession and fed into separate network layers.
• The model was trained and evaluated on its ability to classify publications into UniProtKB categories.

Main Results:

• The CNN model achieved a micro F1-score of 0.72 and a macro F1-score of 0.62.
• This performance significantly outperformed baseline logistic regression and support vector machine models by up to 22 and 18 percentage points, respectively.

Conclusions:

• The proposed CNN model offers a systematic approach for categorizing computationally mapped literature in UniProtKB.
• This method can assist curators in determining publication relevance for protein accession curation.