52/100

Moderate evidence

Cite
More
EmailLinkedInX

AI predicts GenX's odd low-dose effects in zebrafish 91.25% accurately, but only in lab data

Study title: Predicting Hormesis Effects of GenX in Zebrafish via Interpretable Machine Learning: Insights From SHAP Analysis

Evidence type
Benchmark or lab evaluation
Sample
n = 263
Sources
1 sourcePeer-reviewed
Published
1 min read

Study title: Predicting Hormesis Effects of GenX in Zebrafish via Interpretable Machine Learning: Insights From SHAP Analysis

On 2026-10-01, a peer-reviewed study described an interpretable machine learning framework to identify and predict non-monotonic dose-response relationships for GenX, a PFAS alternative, in zebrafish. Using 263 records across 11 biological categories, the CatBoost-based Master Model reached AUC 0.977 and 91.25% accuracy, with SHAP analysis pointing to effect measurement and biological category as key drivers.

Read the study abstract

Hexafluoropropylene oxide-dimer acid (GenX), a prominent alternative to legacy per- and polyfluoroalkyl substances (PFAS), poses a significant challenge to traditional linear risk assessment models due to its ability to induce hormesis-a biphasic "low-dose stimulation, high-dose inhibition" response. This study established an interpretable machine learning (ML) framework to identify and predict GenX-induced non-monotonic dose-response (NMDR) relationships in zebrafish (Danio rerio). By integrating 263 independen…

Predicting Hormesis Effects of GenX in Zebrafish via Interpretable Machine Learning: Insights From SHAP Analysis
Zebrafish Make a Splash in FDA Research (8614532586) by The U.S. Food and Drug Administration. Public domain

The gain

An interpretable machine learning model predicted GenX's low-dose-stimulation, high-dose-inhibition effects in zebrafish with 91.25% accuracy (AUC 0.977) across 263 experimental records.

In brief

The work matters because it shifts ecological risk assessment from empirical curve-fitting to mechanistic predictive generalization for emerging PFAS alternatives, identifying immune, endocrine, reproductive and neurobehavioral endpoints as most sensitive. What remains uncertain is how well predictions transfer beyond zebrafish laboratory data to field ecosystems and human-relevant exposures.

Main points

  1. Study integrated 263 independent experimental records across 11 biological categories to model non-monotonic dose-response relationships.
  2. Benchmarked 6 ML paradigms; CatBoost-based Master Model achieved AUC 0.977 and 91.25% accuracy.
  3. SHAP analysis identified Effect Measurement and Biological Category as primary determinants, with immune, endocrine, reproductive and neurobehavioral endpoints most sensitive to low-dose stimulation.
  4. SAR analysis flagged MolLogP, SlogP_VSA3, VSA_EState series and Chi1v as core structural drivers linked to oxidative stress signaling and metabolic trade-off.

The full story

Researchers compiled 263 zebrafish experimental records spanning 11 biological categories to address GenX-induced hormesis, defined as low-dose stimulation, high-dose inhibition that challenges linear risk models.

The CatBoost-based Master Model outperformed five other ML paradigms, and SHAP interpretation plus SAR analysis linked hormetic sensitivity to specific systems and to lipophilicity, electrostatic and topological descriptors like MolLogP and Chi1v.

Sources

1 on file
  1. IUBMB Lifedoi:10.1002/iub.70135
    Peer-reviewed2026-10-01

Talk it over

Three questions to start on, two more if you have time
The CatBoost-based Master Model reached 91.25% accuracy on 263 zebrafish records from lab experiments. Which kind of result, from which setting, would show whether GenX behaves the same way in the field or for people?

Start with a question

AI predicts GenX's odd low-dose effects in zebrafish 91.25% accurately, but only in lab datahttps://truvace.com/item/predicting-hormesis-effects-of-genx-in-zebrafish-via-interpretable-machine-learn

  1. The CatBoost-based Master Model reached 91.25% accuracy on 263 zebrafish records from lab experiments. Which kind of result, from which setting, would show whether GenX behaves the same way in the field or for people?

    Answer in the thread
  2. The records span 11 biological categories, and SHAP analysis pointed to effect measurement and biological category as key drivers. What might be missing from those records that could change the prediction?

    Answer in the thread
  3. If you were a regulator assessing GenX, how would you weigh a model that predicts low-dose stimulation and high-dose inhibition against the linear risk models it challenges?

    Answer in the thread
If you have more time
  • The model flagged immune, endocrine, reproductive and neurobehavioral endpoints as most sensitive. What new experiment would test whether that holds in species other than zebrafish?

    Answer in the thread
  • The researchers benchmarked 6 ML paradigms. What would you look up to judge whether a 91.25% accuracy figure is high for this kind of problem?

    Answer in the thread

What readers say