Rapid biomarker assay development


We have successfully evaluated a panel of circulatory biomarkers to identify targets that can be used to distinguish disease from non-disease samples.

For this, we have created an algorithm that has the capacity to combine evidence from candidate biomarkers and genetic and other variants to provide a potentially new clinical tool.

In colorectal cancer (CRC) there is a robust signal for a number of biomarkers that is associated to CRC presence or absence. Multivariate classification models were trained and demonstrated a clear ability todiscriminate CRC status, which might be increased with a larger sample size.

In both CRC and BC datasets there seemed to be multiple orthogonal sources of variation in immune system state, so that reducing the biomarkers into fewer variables using principal components would result in loss of potentially valuable immune state variation. Instead the biomarker measurements themselves were used directly in predictive models, and a recursive feature elimination procedure was used to reduce the number of biomarkers to 5, without loss of accuracy in CRC and BC. The models would be improved with more training data, and their performance should be tested and replicated in independent datasets.