Question About PIMMS for Missing Value Imputation

[gaoxiangcao]

Dear Developer,

Can the PIMMS package be used to impute missing values in proteomics datasets processed with MaxQuant, particularly for post-translational modifications? If so, are there specific recommendations for handling left-censored data?

Thank you!

[enryH]

Hey,
yes you can go ahead, there is no particular difference with PTMs. You will just need to have PTM specific quantifications.

Have you seen the tutorial?

Left censored data was in my experiments with relative large and diverse datasets (same cell line measured repeatedly or patient derived samples) not a big issue. I rather found left-censored methods to impute values out of distribution. But it depends a lot on your design. If you have knock-outs, you can basically impute with any value. For differential analysis you do not necessarily have to impute if your groups have similar ratios of missing values for the feature. If you use PIMMS models (CF, DAE, VAE) or other data driven ones (RF, KNN) you would expect to have imputation according to the overall distribution of the feature, probably towards the lower limit of observed intensities.