On 2014 Mar 31, Sergio Stagnaro commented:

None

On 2014 Mar 30, Allison Stelling commented:

The accuracy isn't totally surprising, mass spec would be quite sensitive to pathogenic phenotype markers like those phosopholipids. These smaller biological molecules- lipids, sugars, etc- are linked to the mechanisms of pathogenesis. As biomarkers go, they're pretty solid.

What I worry about is that the algorithm they use to sort the data may have a contaminated "healthy vs normal" training set. I see this in a lot of the spectral diagnostic literature. It is making the assumption that the "gold standard" test for the disease is in fact accurate and reliable itself.

I also see a lot of use of non-open source software in the field. This may hinder our efforts to establish ourselves in the medical realms. We must make these analyses reproducible and reliable across many, many sites for medical work; and that means having many eyes on the code.

On 2014 Mar 27, Robert Tibshirani commented:

This paper shows some impressive results, predicting phenoconversion to either amnestic mild cognitive impairment or Alzheimer's disease, from ten lipids in peripheral blood, with 90-95% accuracy over a 2-3 time frame. The research appears to be well done, but as is often the case, it is impossible to judge the strength of the results without digging into the details. The statistical analysis has multiple steps and is quite complicated. The raw data needs to be made available, and ideally, an R script that carries out the calculations in full.

On 2014 Mar 27, Robert Tibshirani commented:

This paper shows some impressive results, predicting phenoconversion to either amnestic mild cognitive impairment or Alzheimer's disease, from ten lipids in peripheral blood, with 90-95% accuracy over a 2-3 time frame. The research appears to be well done, but as is often the case, it is impossible to judge the strength of the results without digging into the details. The statistical analysis has multiple steps and is quite complicated. The raw data needs to be made available, and ideally, an R script that carries out the calculations in full.

On 2014 Mar 31, Sergio Stagnaro commented:

None