In this work we focus on applying multivariate methods to scanning probe based photothermal nearfield mid-IR spectroscopy (usually referred to as AFM-IR). AFM-IR spectra are often described as resembling bulk absorption spectra. However, in contrast to bulk absorption spectra, AFM-IR spectra do not follow Beer's law and thus their signal does not depend linearly on the analyte concentration. Furthermore, at nanoscale spatial resolution lateral sample drift is noticeably affecting sample positioning. The optical non-linearity can be addressed by sample preparation and careful choice of sampling parameters. Thermal drift is a bit more challenging to solve in practice. Instead of trying to minimize drift beyond what is typically achievable in a conventional AFM-IR setup, we have developed several strategies (measurement procedures and software routines) that allow us to acquire high resolution (both high spatial resolution and high pixel resolution) hyperspectral images in presence of thermal drift. These strategies allow us to apply multivariate methods to AFM-IR hyperspectral images of micro-organisms, vesicles and other chemically complex nanoscale structures whithout loosing information about measurement position and artifacts due to offsets between images.