Raman spectroscopy is widely used in chemistry, material science and in biomedical applications such as cancer detection. A Raman spectrum of tissue shows DNA, amino acids, lipids, and proteins simultaneously, which makes the evaluation of the spectral content both complete but also challenging. The impact of the technique can increase substantially by access to big databases, but variations in the setup, e.g. quantum efficiency of Raman detectors, transmission profiles and disturbances of optical filters and components, may hinder direct data comparison. We here introduce a step prior to preprocessing that calculates spectral quotients to address system-dependent multiplicative differences and system-inherent background noise, thereby enabling analysis of Raman spectral data from different setups. Pre-processing was performed using an artificial neural network (ANN) trained on synthetic data to deal with fluorescent background and noise. Validation by multivariate analysis of the spectral quotients combined with ANN was performed on randomized synthetic data and, as a proof of principle, on experimental data from brain tumor biopsies. The results demonstrated clustering and feature extraction that was not possible without the introduction of the quotients. Data exploration revealed that the method enabled spectral feature identification even for weak Raman signals that are not in resonance with the excitation wavelength. Note, some distortions persist due to data dependency and additive errors but a system independent clustering was achieved. ANN-based preprocessing combined with spectral quotients for combined evaluation of Raman spectroscopic data from multiple setups opens the possibility for more robust multivariate studies in biomedical and other applications.