Protein Annotation Bias Analysis

Introduction

This project analyzes biases in human proteome experimental GO annotations and their correlation with research interest over time (2013-2022).

Dependencies

• Python 3.7+
• R 4.0+
• NCBI EDirect for PubMed data retrieval
• Required Python packages: Bio, obonet, pandas, subprocess, datetime
• Required R packages: tidyverse,reshape2, ineq, ggplot2, ggpubr, cowplot

Data Availability

The data associated with this project is available on Figshare:

• External data files: DOI: 10.6084/m9.figshare.27252282. Download and place the contents in the 0_data/ext_data directory
• Processed data files: DOI: 10.6084/m9.figshare.27252282. RDS files should be placed in the 1_metrics directory to be used for analyses. annotation_pub_years.csv file contains GO annotations (in Decemeber 2022) and the corresponding publication year, which should be placed in 3_delay_curation directory to perform delay in curation analysis.

Usage

0. Data Generation

Knowledge Metrics

Scripts in this group process Gene Ontology Annotation (GOA) files to extract knowledge metrics: term count, unique term count, information content.

• loop_years.py - Iterates through GOA files placed in 0_data/ext_data folder to process each GOA file
• extract_GOA.py - Processes GAF files to calculate GO-based metrics. Information content metric is calculated using pivot year of 2022 (GO structure and GO term frequencies in 2022)
• ia.py - Helper script for Information Accretion calculations
• ia_2022.csv - Information accretion values for all GO terms using 2022 GO structure as a pivot
• process_GOA_years.py - Combines temporal GO-based metrics for each GO aspect and experiment type (articles describing high-throughput or low-throughput experiments)
• ggplot_theme_Publication-2.R - Helper script for ggplot themes in R

Interest Metrics

• process_interest_data.R - Preprocesses data by calculating cumulative full publication equivalents (FPEs), filtering clinical assay proteins (retrieved from MarkerDB), resolving duplicated protein aliases using GO-based metrics. Number of FPEs are obtained from the Pharos/TCRD database, see here.

Knowledge and interest metrics are stored in folder 1_metrics for further analysis. These processed files are available on Figshare (DOI: 10.6084/m9.figshare.27252282)

1. Analyzing Metrics

Scripts in this folder are used to analyze and visualize the calculated knowledge (GO-based) and interest (number of FPEs) metrics of proteins.

• 1.1_all_metrics.R:
  • Compare GO-based metrics of proteins that receive annotations from high-throughput articles and low-throughput articles
  • Examine temporal changes in protein information content and number of FPEs, with an emphasis on most informative proteins
• 1.2_gini.R: Gini coefficient calculations using knowledge and interest metrics, comparing proteins first annotated before 2013 with all proteins

2. Correlation & Disease Association

Scripts in this folder are used to examine correlations between knowledge and interest metrics.

• 2.1_knowledge_vs_interest.R : Knowledge-interest correlations over time with Spearman correlation tests
• 2.2_disease_association.R: Knowledge-interest correlations among proteins associated with diseases (retrieved from DISEASES database)

3. Curation Delay Analysis

Scripts for analyzing the time gap between publication and annotation:

• 3.1_extract_pubyear.py: EDirect-based scripts for retrieving PubMed publication years of the publications recorded in GOA files. Retrieved publication years are merged with GOA file in December 2022, available on Figshare as annotation_pub_years.csv (DOI: 10.6084/m9.figshare.27252282)
• 3.2_delay_analysis.R: Analysis and visualization of temporal differences between publication and annotation dates

Installation

1. Clone this repository
2. Install dependencies
3. Download data from Figshare following the Data Availability section
4. For EDirect data retrieval:

sh -c "$(wget -q https://ftp.ncbi.nlm.nih.gov/entrez/entrezdirect/install-edirect.sh -O -)"
export PATH=${HOME}/edirect:${PATH}

5. Create 4_figures directory to store generated plots. Run the scripts based on the folder order (from 0 to 3).