XNet: A Bayesian Approach to Extracted Ion Chromatogram Clustering for Precursor Mass Spectrometry Data

Liquid chromatography mass spectrometry
is a popular technique for high throughput analysis of biological
samples. Identification and quantification of molecular species via
mass spectrometry output requires postexperimental computational analysis
of the raw instrument output. While tandem mass spectrometry remains
a primary method for identification and quantification, species-resolved
precursor data provides a rich source of unexploited information.
Several algorithms have been proposed to resolve raw precursor signals
into species-resolved isotopic envelopes. Many methods are particularly
dependent on user parameters, and because they lack a means to optimize
parameters, tend to perform poorly. To this end we present XNet, a
parameter-less Bayesian machine learning approach to isotopic envelope
extraction through the clustering of extracted ion chromatograms.
We evaluate the performance of XNet and other prevalent methods on
a quantitative ground truth data set. XNet is publicly available with
an Apache license.