The Effect of Gaussian Noise on Maximum Likelihood Fitting of Gompertz and Weibull Mortality Models with Yeast Lifespan Data

ABSTRACT

Background/study context: Empirical lifespan data sets are often studied with the best-fitted mathematical model for aging. Here, we studied how experimental noises can influence the determination of the best-fitted aging model. We investigated the influence of Gaussian white noise in lifespan data sets on the fitting outcomes of two-parameter Gompertz and Weibull mortality models, commonly adopted in aging research.

Methods: To un-equivocally demonstrate the effect of Gaussian white noises, we simulated lifespans based on Gompertz and Weibull models with added white noises. To gauge the influence of white noise on model fitting, we defined a single index, ${\delta }_{LL}$, for the difference between the maximal log-likelihoods of the Weibull and Gompertz model fittings. We then applied the ${\delta }_{LL}$ approach using experimental replicative lifespan data sets for the laboratory BY4741 and BY4742 wildtype reference strains.

Results: We systematically evaluated how Gaussian white noise can influence the maximal likelihood-based comparison of the Gompertz and Weibull models. Our comparative study showed that the Weibull model is generally more tolerant to Gaussian white noise than the Gompertz model. The effect of noise on model fitting is also sensitive to model parameters.

Conclusion: Our study shows that Gaussian white noise can influence the fitting of an aging model for yeast replicative lifespans. Given that yeast replicative lifespans are hard to measure and are often pooled from different experiments, our study highlights that interpreting model fitting results should take experimental procedure variation into account, and the best fitting model may not necessarily offer more biological insights.

Acknowledgments

The authors appreciate the anonymous reviewers for their valuable comments and suggestions during the revision of this manuscript.

Additional information

Funding

This study was partially supported by the University of Tennessee at Chattanooga start-up fund and by the National Science Foundation Award #1453078 (transferred to #1720215).