Separating the scientists from the folkscientists
http://genomemedicine.com/content/2/2/10/
(emphasis added)
Please note the above article has a real chance of being false.
http://genomemedicine.com/content/2/2/10/
Conclusions
In this paper we set out to compare different models that combine the effects of multiple risk loci into an overall genetic risk. We conclude that a model that is additive or multiplicative on the risk scale across all loci is incompatible with the observed recurrence risks to relatives. The constrained multiplicative (CRisch), Odds and Probit models are all compatible with the observed data and, in fact, it is difficult to distinguish between them when the relative risk at an individual locus is small. Importantly, we show that the unconstrained multiplicative (Risch) model, often used in theoretical studies because of its mathematical tractability, is not a realistic model as impossible probabilities of disease are implied. Specifically, the multiplicative Risch model generates a relationship of = 1, but we have demonstrated that this not possible under many disease scenarios and occurs in the theoretical derivation because probabilities of disease are not constrained and can exceed 1. We have demonstrated that under more realistic models in which probabilities of disease are constrained to 1, the ratio is often much less than 1, a result that is consistent with empirical estimates from a range of diseases. Finally, we conclude that it will only be possible to distinguish between the CRisch, Odds and Probit models in practice if genetic risk profiles are able to reconstruct the majority of the known genetic variance; this is unlikely for the foreseeable future.
In this paper we set out to compare different models that combine the effects of multiple risk loci into an overall genetic risk. We conclude that a model that is additive or multiplicative on the risk scale across all loci is incompatible with the observed recurrence risks to relatives. The constrained multiplicative (CRisch), Odds and Probit models are all compatible with the observed data and, in fact, it is difficult to distinguish between them when the relative risk at an individual locus is small. Importantly, we show that the unconstrained multiplicative (Risch) model, often used in theoretical studies because of its mathematical tractability, is not a realistic model as impossible probabilities of disease are implied. Specifically, the multiplicative Risch model generates a relationship of = 1, but we have demonstrated that this not possible under many disease scenarios and occurs in the theoretical derivation because probabilities of disease are not constrained and can exceed 1. We have demonstrated that under more realistic models in which probabilities of disease are constrained to 1, the ratio is often much less than 1, a result that is consistent with empirical estimates from a range of diseases. Finally, we conclude that it will only be possible to distinguish between the CRisch, Odds and Probit models in practice if genetic risk profiles are able to reconstruct the majority of the known genetic variance; this is unlikely for the foreseeable future.
Please note the above article has a real chance of being false.
Comment