From expansion to consolidation: two decades ofGene Ontology evolution

This study presents a 21-year temporal analysis of the Gene Ontology and its annotations, revealing a shift from sustained growth and structural reorganization to a maturation phase of increased stability beginning around 2017, thereby providing a critical framework for ensuring the reproducibility and long-term reuse of GO-based research.

The Gist

Imagine the Gene Ontology (GO) as the world's most massive, living library catalog for biology. Instead of organizing books, it organizes the functions of every gene in every living thing. If you want to know what a specific gene does (like "helps the heart beat" or "digests sugar"), you look it up in this catalog.

This paper is like a 21-year history book of that library, written by researchers who watched the catalog grow from 2004 to 2024. They wanted to answer a simple question: How has this catalog changed over time, and does it matter if we use an old version or a new one?

Here is the story of the paper, broken down with some everyday analogies:

1. The Library is Growing Up (From Expansion to Consolidation)

For the first 15 years, the library was in a phase of wild expansion.

• The Analogy: Imagine a new town being built. In the beginning, people are just throwing up houses, roads, and parks everywhere. It's chaotic but fast. New terms (words describing gene functions) were being added every single day.
• The Shift: Around 2017, the town planners realized, "Wait, we have enough houses. Let's stop building new ones and start fixing the old ones."
• The Result: The paper calls this a transition from expansion to consolidation. The library stopped adding massive amounts of new shelves and started reorganizing the existing ones. They began retiring old, confusing labels (obsolete terms) and making the structure cleaner. It's like a teenager growing up and becoming a responsible adult; the resource is maturing.

2. The Three Departments (BP, MF, CC)

The library is split into three main departments, each with its own personality:

• Biological Process (BP): The "What is happening?" department (e.g., "cell division"). This is the biggest and most complex section.
• Molecular Function (MF): The "What can it do?" department (e.g., "cuts DNA").
• Cellular Component (CC): The "Where is it?" department (e.g., "inside the nucleus").

The Discovery: The "What is happening?" department (BP) was the most chaotic. For years, they kept adding new "middle managers" (general terms) to the hierarchy, making the structure wider but not necessarily deeper. Around 2017, they finally stopped adding so many middle managers and started straightening out the top levels of the organization chart.

3. The "Top Floor" Shake-up

One of the most interesting findings was about the top floor of the library (the most general terms).

• The Analogy: Imagine the top floor of a hotel. Usually, you expect the lobby and the main elevators to stay the same forever. But in this library, the lobby got completely redesigned around 2018.
• Why it matters: If you are a researcher looking for a gene, and the "Lobby" (the main categories) has moved, you might get lost. The paper shows that these big, structural changes happened mostly in the last decade, meaning the "map" of biology is being redrawn at the highest levels to make more sense.

4. The Annotations (The Book Reviews)

The catalog (the terms) is useless without the annotations (the actual notes linking genes to those terms). The researchers looked at three different "librarians" who write these notes:

• SGD (Yeast): A very careful, manual librarian who has been working for decades. Their notes are high-quality and stable.
• MGI (Mouse): A busy librarian covering a complex organism. Their notes grew steadily as more experiments were done.
• GOA (UniProt): A massive, automated robot librarian that covers every species.
  • The Twist: The robot librarian (GOA) used to rely heavily on "electronic inference" (guessing based on patterns). The paper noticed that around 2018, the robot changed its software. Suddenly, the number of "guessed" notes stabilized, and the way they were generated changed. This shows that even automated systems have "software updates" that change the data you see.

5. Why Should You Care? (The "Time Travel" Problem)

This is the most important part for anyone using this data.

• The Problem: If you run a computer analysis on gene data using the 2010 version of the catalog, you might get a different answer than if you run it with the 2024 version.
• The Analogy: It's like using a GPS. If you use a map from 2010, it might tell you to turn left at a street that was closed in 2015. If you use the 2024 map, it routes you correctly.
• The Takeaway: The paper warns scientists: "Always check your map version!" Because the catalog changes, your results are tied to the specific year you used. To make science reproducible (so others can repeat your work), you must say exactly which version of the Gene Ontology you used.

Summary

The Gene Ontology has gone through a 20-year journey:

1. Childhood (2004–2016): Rapid growth, adding new terms and expanding the structure.
2. Adulthood (2017–Present): Maturation. The growth slowed down, the structure was reorganized to be clearer, and the system became more stable.

The paper tells us that while the library is still changing, it is becoming a more reliable, well-organized place. However, because it does change, scientists need to be careful to note which "edition" of the library they are using, or they might end up reading the wrong story.

Technical Summary

1. Problem Statement

The Gene Ontology (GO) is the foundational resource for functional annotation of gene products across biological databases. However, GO is a dynamic, continuously evolving resource. Its structure (terms and relationships) and associated annotations change over time due to new biological discoveries and curation efforts.

• The Challenge: Analytical results derived from GO (e.g., enrichment analysis) are inherently tied to specific ontology and annotation versions. Without a systematic understanding of how GO changes, data reuse, benchmarking, and reproducibility are compromised.
• The Gap: While previous studies have examined limited time spans or specific aspects of GO evolution, there is a lack of a comprehensive, long-term characterization (spanning two decades) that treats successive releases as a unified longitudinal dataset, covering both the ontology structure and annotation content.

2. Methodology

The authors conducted a large-scale temporal analysis of GO releases spanning 21 years (2004–2024). The study was divided into two main components:

A. Data Sources

• Ontology Data: Retrieved go.obo files from the GO Archive for every year from 2004 to 2024.
• Annotation Data: Analyzed annotations from three distinct resources representing different curation models:
  1. SGD: Saccharomyces cerevisiae (Yeast) – High manual curation, deep study.
  2. MGI: Mus musculus (Mouse) – Complex multicellular model, heterogeneous evidence.
  3. GOA-UniProt: Broad taxonomic coverage, heavy reliance on automated pipelines (IEA).
     Note: WormBase data for C. elegans was also mentioned in the text as part of the analysis scope.

B. Analytical Framework

The analysis was stratified by the three GO subontologies: Biological Process (BP), Molecular Function (MF), and Cellular Component (CC).

1. Ontology Structure Metrics:

   • Term Dynamics: Tracked active vs. obsolete terms, new term additions, and obsoletion rates.
   • Structural Topology: Analyzed the ratio of leaf terms (most specific) to non-leaf terms (general/internal nodes).
   • Depth Analysis: Calculated the median depth of terms (distance from the root) to determine if the ontology was growing "deeper" (linear extension) or "broader" (lateral expansion).
   • Relationships: Grouped relationships into is_a, part_of, and "other" types to track structural density changes.
   • First Layer Analysis: Monitored the size of the top-level terms (direct children of the root) to detect major structural reorganizations.

2. Content Analysis:

   • Word Enrichment: Performed hypergeometric survival function tests on newly added term names and synonyms for each year to identify shifting biological themes (e.g., specific organ systems or diseases).

3. Annotation Dynamics:

   • Tracked total annotation counts and stratified them by evidence codes (Experimental, Phylogeny, Computational, Author, Curator, and Inferred from Electronic Annotation - IEA).
   • Compared growth trajectories across the different databases (SGD, MGI, GOA).

3. Key Results

A. Ontology Evolution: From Expansion to Consolidation

• Growth Trend: All three subontologies showed steady growth in active terms until ~2017, after which the growth plateaued.
• The "Tipping Point" (2017–2019): A distinct transition occurred around 2017.
  • Additions: The rate of new term additions declined significantly across all subontologies.
  • Obsoletion: The rate of term obsoletion increased, particularly in BP and CC.
  • Stability: The resource entered a phase of "consolidation" or maturation, characterized by increased stability and incremental rather than disruptive changes.
• Structural Reorganization:
  • BP (Biological Process): Prior to 2017, BP expansion was driven by the addition of internal (non-leaf) nodes, leading to a "broader" ontology rather than a "deeper" one (median depth actually decreased until ~2011). Post-2017, the ratio of non-leaf to leaf terms stabilized.
  • First Layer Volatility: The top-level terms (closest to the root) showed significant fluctuations, with major reorganizations occurring around 2018 in MF and CC. This indicates deep conceptual revisions in how core biological functions are classified.
  • Relationships: The ratio of is_a and part_of relationships showed abrupt changes in CC around 2019, consistent with known large-scale restructuring efforts (e.g., CARO principles).

B. Thematic Shifts

• Word enrichment analysis revealed that the biological focus of new terms shifts with research priorities.
  • 2009: Enrichment in gland-related terms.
  • 2010: Kidney-related terms.
  • 2016: Significant additions related to the nervous system (linked to Parkinson's disease research efforts).

C. Annotation Dynamics

• Divergent Growth Patterns:
  • SGD (Yeast): Moderate growth 2004–2007, sharp expansion in 2008, followed by gradual growth and recent stabilization.
  • MGI (Mouse): Steady increase across all aspects, with a local peak in 2007.
  • GOA (UniProt): Two distinct phases: steady growth (2010–2017), a sharp rise in 2018, and moderate growth thereafter.
• Evidence Code Shifts:
  • Experimental Annotations: Increased steadily in SGD and MGI but remained constant in GOA.
  • IEA (Electronic Annotations): Showed high variability. GOA saw a massive introduction of IEAs starting in 2018. MGI showed a peak in 2007 followed by a decline.
  • Phylogeny-based Annotations: Notable decline in GOA starting in 2019.

4. Key Contributions

1. Longitudinal Dataset Characterization: Provided the first comprehensive 21-year temporal analysis of GO, treating ontology and annotations as a unified, evolving research dataset.
2. Identification of a Maturation Phase: Identified a critical transition point around 2017, marking a shift from rapid expansion to structural consolidation and increased stability.
3. Structural Insight: Demonstrated that GO:BP growth was primarily lateral (broadening) rather than linear (deepening), and highlighted that the most volatile changes occur at the highest levels of the ontology hierarchy.
4. Resource-Specific Dynamics: Mapped how different curation strategies (manual vs. automated) and organism scopes influence annotation growth and evidence composition.

5. Significance and Implications

• Reproducibility: The study underscores that GO-based analyses are time-dependent. Researchers must explicitly report the specific GO version and annotation release used to ensure reproducibility.
• Tool Development: Bioinformatics tools must be designed to handle versioning explicitly (FAIR principles). Systems should track ontology and annotation versions to allow for consistent reanalysis of historical data.
• Resource Maturity: The identification of the 2017 "tipping point" suggests that GO has reached a mature state where future changes will be more incremental. This provides confidence for long-term data integration projects but also highlights the need for careful version management during the transition period.
• Interpretation of Past Studies: Historical analyses should be re-evaluated in the context of the ontology state at the time of generation, as structural reorganizations (especially in the upper layers) can alter biological interpretations.

In summary, the paper establishes that while GO has grown significantly over two decades, it has recently entered a phase of consolidation. Understanding these temporal dynamics is crucial for the accurate, transparent, and reproducible use of GO in modern computational biology.