US HUPO 2019

Was this article helpful?

Model-Free SILAC Data Analysis is Possible, Reproducible, and Essential

David Chiang; Patrick Chu
Sage-N Research, Milpitas, CA

ABSTRACT:  (View the Poster Here)

Using SILAC as a case study, we aim to solve this paradox: How proteomics — an analytical science with high-accuracy data — can produce irreproducible results.

One factor is that labs rely on complex models (search engines, discriminant scores, posterior probabilities) within often-opaque software to interpret m/z data. Increased complexity generally means narrower applicability. Because mass spectrometry data have wide signal-to-noise, any complex model is prone to fitting problems for some data subset while depriving researchers of data insight.

For SILAC differential quantitation, a conventional workflow uses a statistical formula to calculate a “probability” (actually p-value) to evaluate a peptide ID, then fit a Gaussian area-under-curve model to quantify light and heavy peptides to derive their ratio. It’s simple in concept but tricky in practice.

Experienced users often find that while 90% or more of the protein ratios are generally correct (if imprecise), a small portion (but often the most critical) have essentially random ratios.

Two susceptibilities explain this: (1) Over-fitted p-values randomizes IDs and (2) under-fitted areas can randomize quantitation. A Gaussian fit requires at least one data-point near the maximum. In any large dataset, some small percentage would lack such points to yield random ratios.

Counter to complexity bias, a robust foundation requires direct analysis of raw m/z data with model-free peptide significance and SILAC quantitation.

Specifically, each peptide-sequence match (PSM) can be distilled to (1) number of matched fragments and (2) their average RMS error. For SILAC quantitation, each precursor scan is essentially an independent ratio-sampling experiment whose median approximates the true ratio. We believe this model-free approach is novel and fundamentally solves irreproducibility.