How is inTrigue different?
inTrigue does not look at diagnosed patients and create simple similarity searches. It does not seek to combine existing data sets to create symptom lists or genome mutation libraries. None of these approaches work effectively, and all produce poor results with many false positives. Precision is low, when we need precision to be high.
It does not search by genotype. For many rare diseases we have worked on, we have found that gene mutations have insufficient predictive power to indicate symptom onset or severity.
inTrigue does not do simple search by collections of phenotypes. For many diseases, this approach fails to account for the differences in the way in which patients are managed in different healthcare systems, and it does not work. Stated differently, simple algorithms learnt like this in one country cannot be ported unchanged to another country.
It does not need any positive class patients (with disease) to build the disease models. Looking for undiagnosed patients is a discovery process and using other methods will not discover the hidden patients.
inTrigue does not seek to remove features that are sparse, as it can manage sparsity in the data set.
That's amazing, and it is the way to better results.