NIMBus: a negative binomial regression based Integrative Method for mutation Burden Analysis

摘要

Flowchart of NIMBus. For a given disease d(1≤d≤D), sd represents the total number of samples for that disease. In addition, there are a total of m features which are denoted as f1⋯fm. The mutations from the samples and the features are binned on the bins b1⋯bn for a total of n. Two resulting matrices are produced, Y and X. The matrix Y is a D×n matrix consisting of mutation counts while X is an n×m matrix consisting of feature values. Training the negative binomial model gives, for each disease, μ and σ values for each bin, n. The trained model can be applied to a set of user defined regions, 1⋯K, to evaluate relative mutation burden. This testing is associated with a set of P-values, p, for each of the K regions. The P-values from multiple regions may be combined using Fisher’s method