Paper Summary

Citation: Dhaked, D. K., Ihlenfeldt, W.-D., Patel, H., Delannée, V., & Nicklaus, M. C. (2020). Toward a Comprehensive Treatment of Tautomerism in Chemoinformatics Including in InChI V2. Journal of Chemical Information and Modeling, 60(3), 1253–1275. https://doi.org/10.1021/acs.jcim.9b01080

The Problem: Chemical Identifiers That Miss the Mark

Imagine you’re managing a chemical database with millions of compounds. You encounter two entries: one for keto-glucose and another for enol-glucose. To a human chemist, these are obviously the same molecule—just different tautomeric forms that rapidly interconvert in solution. But to your database? They’re completely different compounds with different identifiers, different entries, and different search results.

This is the tautomerism problem in chemoinformatics, and it’s far more pervasive than you might think. Even our most sophisticated chemical identifiers, including the International Chemical Identifier (InChI), struggle to recognize when two molecular representations are actually the same dynamic molecule existing in different forms.

What Makes Tautomerism So Tricky?

Tautomerism—the rapid interconversion between structural forms through hydrogen atom movement—presents unique challenges for computational chemistry:

Context Dependency: Unlike fixed molecular properties, tautomeric preferences depend heavily on conditions like temperature, pH, and solvent. A molecule might prefer its keto form in water but its enol form in a non-polar solvent.

Speed vs. Accuracy: With databases containing hundreds of millions of structures, we can’t run quantum mechanical calculations to predict every tautomeric relationship. We need fast, rule-based approaches that can process massive datasets in reasonable time.

Database Chaos: Current chemical identifiers create “tautomeric conflicts”—the same molecule listed multiple times under different identifiers simply because it can exist in different tautomeric forms.

A Comprehensive Rule Set: 86 Ways Molecules Transform

The researchers tackled this problem systematically, developing 86 tautomeric interconversion rules based on experimental literature. This isn’t just an academic exercise—these rules were designed to inform the development of InChI Version 2.

The Rule Categories

The 86 rules break down into three main types:

  • 54 Prototropic Rules: Classic tautomerism involving hydrogen movement, like keto-enol interconversions
  • 21 Ring-Chain Rules: Transformations between cyclic and open-chain forms
  • 11 Valence Rules: Structural rearrangements involving valence changes

Testing at Scale: 400 Million Molecules

Here’s where the research gets impressive. The team didn’t just propose these rules—they tested them against nine major chemical databases totaling over 400 million structures. This massive analysis revealed some fascinating patterns.

The most common rule, PT_06_00 (covering most keto-enol cases), applies to over 70% of molecules in the combined databases. But the analysis also uncovered that many newly defined rules affect millions of compounds that current InChI versions completely miss.

The Current InChI Problem

The results reveal a significant gap in current chemical identification systems. Even the most flexible “Nonstandard” InChI settings only achieve about 50% success in recognizing tautomeric relationships. Some crucial new rules show success rates below 2%.

This isn’t just a technical curiosity—it has real-world implications:

  • Database Redundancy: The same compound appears multiple times under different identifiers
  • Search Failures: Researchers miss relevant compounds during database searches
  • ML Training Issues: Machine learning models trained on these databases learn to treat tautomers as different molecules

What This Means for InChI V2

The research suggests that implementing this comprehensive rule set would triple the number of compounds affected by tautomerism recognition. This necessitates a major redesign for InChI Version 2, not just incremental improvements.

The implications are significant for anyone working with chemical databases:

  • Better Data Quality: Reduced redundancy and improved consistency in chemical databases
  • Enhanced Search: More comprehensive results when searching for molecular structures
  • ML Applications: More accurate training data for machine learning models in drug discovery and molecular property prediction

Practical Applications: The Tautomerizer Tool

Recognizing that chemists need practical ways to explore these concepts, the researchers created Tautomerizer, a public web tool that allows users to test these rules on their own molecules. This demonstrates the practical focus of the work—not just theoretical rule development, but usable tools for the chemistry community.

Looking Forward

This research represents more than just an academic study—it’s laying the groundwork for the next generation of chemical informatics tools. By providing a comprehensive, experimentally-grounded set of rules, the work enables more accurate molecular databases, better search algorithms, and more reliable machine learning models.

For researchers working at the intersection of chemistry and data science, this represents a crucial step toward chemical identifiers that actually capture the dynamic nature of molecular reality.

References

  • Dhaked, D. K., Ihlenfeldt, W.-D., Patel, H., Delannée, V., & Nicklaus, M. C. (2020). Toward a Comprehensive Treatment of Tautomerism in Chemoinformatics Including in InChI V2. Journal of Chemical Information and Modeling, 60(3), 1253–1275. https://doi.org/10.1021/acs.jcim.9b01080