Phenomenological Load on Model Parameters Can Lead to False Biological Conclusions

Area of Science:

• Evolutionary biology
• Bioinformatics
• Computational biology

Background:

• Maximum likelihood fitting of substitution models aims to explain site-pattern variation.
• Parameters can absorb variance (phenomenological load) even if not mechanistically relevant.
• Phenomenological load can inflate statistical significance and obscure true evolutionary processes.

Purpose of the Study:

• To investigate phenomenological load in substitution models.
• To identify conditions under which phenomenological load arises.
• To develop a method for detecting true evolutionary signals despite phenomenological load.

Main Methods:

• Utilized a mutation-selection framework to generate realistic data mimicking mammalian mitochondrial DNA alignments.
• Simulated data to explore phenomenological load under specific conditions.
• Developed and tested a method to identify signals of underlying processes.

Main Results:

• Phenomenological load occurs when substitution models are underspecified and parameters are confounded with data-generating processes.
• The study demonstrates conditions leading to significant phenomenological load.
• A novel method was presented to disentangle true signals from phenomenological load.

Conclusions:

• Phenomenological load is a critical issue in phylogenetic model fitting, especially with realistic data.
• Understanding and addressing phenomenological load is essential for accurate evolutionary inference.
• The proposed method aids in robustly identifying evolutionary signals in complex datasets.