Develop a set of clinical markers quickly, by assessing a large population of literature suggested targets.
187 candidate marker proteins
337 peptides monitored
674 monitored transitions
30 minute LCMS
274 patient blood plasma samples
137 biopsy-confirmed cancer
137 age- gender-matched controls
0.91 ROC AUC
Efforts in discovering biomarkers from liquid biopsies with sufficient performance to be clinically useful remain a significant challenge in diagnostics. However, as detailed in this example, researchers can leverage highly multiplexed approaches to simultaneously measure a multitude of proteins. Specifically, targeted liquid chromatography coupled mass spectrometry (LCMS) using multiple reaction monitoring (MRM) demonstrates how such an approach can result in the discovery of candidate colorectal cancer biomarkers by screening targets suggested by literature. Starting from mined literature among studies exploring protein and genetic factors putatively associated with colorectal cancer, 187 proteins were selected for assay development. Evaluated in a case-control study design with 274 samples the efforts achieved a validation ROC AUC with 87% sensitivity and 81% specificity.
Chromatographic drift was assessed to tolerate approximately 100 LCMS injections between chromatographic column exchanges which were triggered when the lower 97.5 quartile in deviation from the expected retention time was within 21 seconds.
Overall performance of the MRM assay estimated that for 424 transitions, represented by 260 peptides from 168 proteins, responses were quantitatively measured with over 90% confidence. The median coefficient of variation for both instruments was 0.214 and 0.228. Dynamic range was determined to be approximately 2.5 to 3 orders of instrument magnitude, with good linearity between both instruments.
Using one half of the data as a discovery set (69 disease cases and 69 control cases), the elastic net feature selection and random forest classifier assembly were used in cross-validation to identify a 15-transition classifier. The mean training receiver operating characteristic area under the curve (ROC AUC) was 0.82. After final classifier assembly using the entire discovery set, the 136- sample (68 disease cases and 68 control cases) validation set was evaluated. Here, the validation ROC AUC was 0.91. At the point of maximum accuracy (84%), the sensitivity was 87% and the specificity was 81%.
The total assay development time was reported as 2 months with the greatest expense tied to synthetic analyte controls. Given the nature of this approach and subsequent success we expect more clinically oriented evaluations to follow.