I'm largely leaving this one for me, but for proteomics people take a think about how value neural networks (like DIA-NN) have helped us in distinguishing signal from noise in low level peptide data.
They really are evolving. We constantly do true/false by lying to the neural networks about where and when we have blanks and controls and non-human samples. And even 3 years ago, they weren't nearly as trustworthy as they are now.
Now...consider the metabolomics field which doesn't even have a good way for routine false discovery rate estimation in their data....then drop your signal down to just above your noise....
I'm clearly not qualified to really review this study, but I did enjoy it and it made me worry about some upcoming work we have intended to target peptides from single cells where we won't benefit from finely tuned machine learning/neural-networks to tell true from false....Makes me think we'll need to be even more skeptical of each peptide retention time, ion mobility and fragment distribution pattern than usual....
No comments:
Post a Comment