Phenomenological Load on Model Parameters Can Lead to False Biological Conclusions.
Mol Biol Evol
; 35(6): 1473-1488, 2018 06 01.
Article
en En
| MEDLINE
| ID: mdl-29596684
ABSTRACT
When a substitution model is fitted to an alignment using maximum likelihood, its parameters are adjusted to account for as much site-pattern variation as possible. A parameter might therefore absorb a substantial quantity of the total variance in an alignment (or more formally, bring about a substantial reduction in the deviance of the fitted model) even if the process it represents played no role in the generation of the data. When this occurs, we say that the parameter estimate carries phenomenological load (PL). Large PL in a parameter estimate is a concern because it not only invalidates its mechanistic interpretation (if it has one) but also increases the likelihood that it will be found to be statistically significant. The problem of PL was not identified in the past because most off-the-shelf substitution models make simplifying assumptions that preclude the generation of realistic levels of variation. In this study, we use the more realistic mutation-selection framework as the basis of a generating model formulated to produce data that mimic an alignment of mammalian mitochondrial DNA. We show that a parameter estimate can carry PL when 1) the substitution model is underspecified and 2) the parameter represents a process that is confounded with other processes represented in the data-generating model. We then provide a method that can be used to identify signal for the process that a given parameter represents despite the existence of PL.
Texto completo:
1
Colección:
01-internacional
Banco de datos:
MEDLINE
Asunto principal:
Selección Genética
/
Mutación Silenciosa
/
Mamíferos
/
Modelos Genéticos
/
Mutación
Tipo de estudio:
Qualitative_research
Límite:
Animals
Idioma:
En
Revista:
Mol Biol Evol
Asunto de la revista:
BIOLOGIA MOLECULAR
Año:
2018
Tipo del documento:
Article
País de afiliación:
Canadá