Analysis of Independent Differences (AID) detects complex thermal proteome profiles independent of shape and identifies candidate panobinostat targets

Abstract

Identifying global cellular targets of small molecules is a challenge for drug discovery. Thermal proteome profiling (TPP) is a recent technique that uses quantitative proteomics to identify all small molecule protein targets in a single experiment. One current TPP analysis method relies on two major assumptions: sigmoidal melting curve behavior and that intra-condition dependencies preclude an independent and identically distributed model. Herein, we use a previously published panobinostat TPP dataset to show that these assumptions do not hold true and present a novel, shape-independent method, named Analysis of Independent Differences (AID). For each temperature, AID models the differences between conditions of fractions of non-denatured protein as an independent Normal distribution, resulting in a Multivariate Normal observation for each protein. The log of a Multivariate Normal p-value ranks the proteins from most to least likely shifted, and individual Normal p-values within each protein allow for qualitative inspection. Applying AID to the panobinostat dataset revealed known targets in the top 3% of most shifted proteins, as well as candidate targets involved in myeloid leukocyte activation. AID detects complex melting profiles and can be extended to any number of temperature channels, ligand-protein or protein-protein interactions, or general curve data for deeper biological insight.