GSoC Week8

 

Background & Problem

Insufficient human-labeled EC number → SBO term pairs limit downstream machine learning (ML) training effectiveness and generalization.

Solution

Leverage GPT-5’s biochemical knowledge to generate 8,000 high-quality EC–SBO pairs as data augmentation corpora to boost downstream model performance.

Architecture

  • Model choice: Use GPT-5 (stronger reasoning) instead of GPT-4.

  • Structured output: A unified JSON schema to ensure parsability and consistency.

Prompt Engineering

  • Inject biochemical rules (e.g., NAD⁺/NADH redox patterns).

  • Leaf-first strategy in the SBO ontology to avoid over-generalization.

  • Candidate list constraint: force selection from a provided list to reduce hallucinations.

Quality Control

  • Require a reason field to enable consistency review.

  • Extract keywords to aid subsequent rule/retrieval checks.

  • Error handling: format validation, retry/backoff, and isolation of problematic samples to keep the pipeline stable.

Value (Data Augmentation)

  • Larger training set: 8,000 high-quality pairs can significantly improve classification/retrieval models.

  • Long-tail coverage: fills labeling gaps for rare EC numbers.

  • Cost-effective: compared to manual labeling, AI-assisted generation is faster and cheaper.

This Week’s Outputs

  • Completed batch generation and auto-QC for 8,000+ EC–SBO pairs.

  • Finalized structured output and candidate-constraint templates, producing reusable production scripts.

Comments

Post a Comment

Popular posts from this blog

GSoC Week1

Hi, This week focused on the design and implementation of a dynamic update mechanism for SBO (Systems Biology Ontology) data. Key Decisions and Progress 1. Data Update Strategy Decision : On application startup → Check for updates → Download new data → Application becomes available Reason : Prioritize data consistency Status : ✅ Implemented and Pull Request submitted 2. User Control Over Updates Draeger’s Suggestion : Allow users to skip updates to support reproducibility of specific versions Nantia’s  Suggestion : Enforce updates to ensure data consistency Status :  Further discussion and decision in Wednesday's meeting 3. Data Version Management Strategy System Version History Management strategy1 :  Only retain the latest version to save storage  startegy 2:Retain up to 3 versions (2 historical + 1 latest) Status :  Further discussion and decision in  Wednesday's meeting User Import History Management strategy：...
Read more

GSoC week2

Hi, As discussed in the last blog, I started working on my first milestone of fetching SBO table from OLS to SBO annotator. This blog contains a summary of the meeting with the new features I added and the issues I worked on and fixed this week to add SBML L3V2 support. So, let's quickly dive in summary: 1. Change Detection Mechanism for SBO Terms ❓ Issue : OLS API’s updated field changed, but no observable differences in content. ✅ Observation : Compared OLS (API) and GitHub .owl versions – content is identical . Latest update timestamp on OLS is 2025, but GitHub file was last modified in 2023. New terms are already included in the GitHub version. ✅ Decision : We can rely on GitHub .owl file for consistency and reduced request time. 2. Fetching ‘is_a’ / Parent Info Performance Issue ⏱️ Fetching subclass_of (parent info) via OLS takes ~10 minutes. ⚡ Other fields can be fetched in <1 minute. ✅ Optimization : Parent info should be pulled from ...
Read more

GSoC Week 5

 This week, I mainly focused on diagnosing why adding KEGG still failed to yield finer-grained SBO classifications across 108 models, extending the EC lookup pipeline, and experimenting with additional data sources. Here’s the complete summary of the work update during the week. Root-Cause Investigation Added adviaecnumber() helper to attempt refinement via EC number –driven mapping. Re-ran the enhanced pipeline on 108 models  ，reactions that remained in generic SBO classes were still  missing EC numbers . Result: no additional reactions could be refined to more specific SBO classes. Conclusion: the main blocker is not the mapping rule itself but the absence of EC numbers for those reactions across the currently consulted sources. New Data Sources Added SEED adapter : uses the Solr query interface with strict null filtering to avoid spurious entries. Reactome adapter : combines web parsing with the QuickGO API to perform a multi-step conversion...
Read more