The Reality of Ready Biodegradation – New Benchmarks for In Silico Models

When we discuss the environmental risk assessment of industrial chemicals, Ready Biodegradation (RB) is often our first line of defense. However, for many of the thousands of chemicals registered under REACH, experimental data remains a bottleneck. While in silico tools like Biowin, Opera, and Vega promise to fill these gaps, how much can we actually trust them?

A recent landmark study by Karamertzanis et al. (2026) provides a much-needed reality check. By leveraging a curated dataset of 2,684 REACH substances, the authors conducted an unbiased external validation of our industry’s go-to QSAR models.

Key Insights from the Study:

• The 75% Ceiling: While literature often claims accuracies above 85%, this independent assessment found that for industrial chemicals, balanced accuracy plateaus around 0.75.

• The Legend of Biowin 6: Surprisingly, the "venerable" Biowin 6 (based on group contribution methods) remains one of the top performers. Its ability to account for the number of fragments and rings, rather than just their presence, gives it an edge over some modern binary fingerprint models.

• The Consensus Paradox: Using multiple models (majority voting) didn't significantly boost accuracy. It did increase confidence in "Non-Readily Biodegradable" predictions, but at the cost of a much narrower Applicability Domain (AD).

• The Experimental "Speed Limit": Experimental variability in OECD 301 tests effectively sets a "performance ceiling" for any model at approximately 96%. We aren't there yet.

What This Means for Your Risk Assessment

If you are using these tools for regulatory submissions, the study highlights a critical "Negative Predictive Value" (NPV) issue. A prediction of "Ready Biodegradable" is only correct about 34% to 71% of the time depending on the model. This means that using a single model to waive a test is still a high-risk strategy without expert interpretation.

How ToxNavigation Can Help

Navigating these "applicability domains" requires more than just pushing a button. At ToxNavigation, we bridge the gap between software output and regulatory acceptance:

1. Consultancy: We help you build robust, weight-of-evidence cases, selecting the right battery of models to maximize predictive reliability.

2. Training: Our eLearning and tutor-assisted courses are designed specifically for toxicologists who want to master in silico workflows without becoming computer scientists.

Read the full open-access paper:

P.G. Karamertzanis, H. Ekholm, A. Yli-Tuomola, R. Cesnaitis, K. Andreou, A.-M. Nyman, W.D. Coen, A comparative assessment of predictive methods for ready biodegradation using REACH experimental data, Computational Toxicology 37 (2026) 100398. https://doi.org/10.1016/j.comtox.2025.100398.