Support vector regression and least squares support vector regression for hormetic dose-response curves fitting.

Area of Science:

• Toxicology
• Computational Biology
• Environmental Science

Background:

• Hormetic dose-response curves (DRC) are critical for assessing pollutant efficacy and hazards.
• Accurate curve fitting is essential for understanding hormetic phenomena.
• Traditional methods often require large sample sizes, which are not always feasible in experimental toxicology.

Purpose of the Study:

• To evaluate the efficacy of Support Vector Regression (SVR) and Least Squares Support Vector Regression (LS-SVR) for fitting hormetic dose-response curves (DRC).
• To demonstrate the applicability of SVR and LS-SVR for small sample datasets common in experimental toxicology.
• To validate the models using both internal and external validation techniques.

Main Methods:

• Support Vector Regression (SVR) and Least Squares Support Vector Regression (LS-SVR) were employed.
• Model tuning parameters (C, p1 for SVR; gam, sig2 for LS-SVR) were optimized.
• Internal validation used Leave-One-Out (LOO) cross-validation.
• External validation involved splitting the dataset into training and test sets.

Main Results:

• SVR and LS-SVR accurately described the J-shaped hormetic DRC of seven water-soluble organic solvents.
• The models also accurately fitted the classical sigmoid DRC of six pesticides.
• Both methods proved effective for small sample sizes in experimental toxicology.

Conclusions:

• SVR and LS-SVR are effective tools for accurately describing hormetic dose-response curves (DRC).
• These methods offer a robust solution for curve fitting in experimental toxicology, particularly with limited data.
• The findings support the use of SVR and LS-SVR for a better understanding of pollutant effects exhibiting hormesis.