ProteoCure Webinar: Dr. José M. Padrón: “Early pharmacological profiling by label-free continuous live cell imaging”

The identification of drug targets remains a major challenge in drug discovery.
The challenge in a phenotypic drug discovery strategy is to fully understand and elucidate the mode of action, in addition to the resolution of the mechanism of action (MoA), identifying the molecular target(s) affected(s) by the molecule and responsible for its pharmacological activity. Fortunately, this complexity reduces substantially with the development of advanced testing technologies and bioinformatics tools that make phenotypic drug discovery an accessible and realistic strategy. Despite this, there is no established general methodology and the scientific literature on this transcendental topic is scarce and atomized. Studies concerning the in silico prediction of MoA to date rely on data obtained from costly experimental measurements.
In this work, we explored the combination of label-free continuous live-cell imaging and machine learning techniques as a promising tool to depict in a fast and affordable test the mode of action of small molecules with antiproliferative activity. To develop the model, we selected the non-small cell lung cancer cell line SW1573, which was exposed to the known antimitotic drugs paclitaxel, colchicine and vinblastine. The novelty of our methodology focuses on two main features with the highest relevance, (a) meaningful phenotypic metrics, and (b) fast Fourier transform (FFT) of the time series of the phenotypic parameters into their corresponding amplitudes and phases. The resulting algorithm was able to cluster the microtubule disruptors, and meanwhile showed a negative correlation between paclitaxel and the other treatments. The FFT approach was able to group the samples as efficiently as checking by eye. This methodology could easily scale to group a large amount of data without visual supervision.