Force curve classification using independent component analysis and support vector machine

The development of single-molecule force spectroscopy (SMFS) technique, especially the atomic force microscope (AFM) based SMFS technique, has been widely applied to the studies of receptor-ligand at single-cell and single-molecule level and has greatly enhanced the understanding of biological activity like the drug action on the cells. The studies have shown that three types of acting forces between proteins and ligands, specific binding, non-specific binding, and non-interaction, can be distinguished manually according to the characteristics of force curves for further analysis. However the efficiency of manual classification of such force curves is low and results in difficulty in analyzing large set of experimental data. In this study, we demonstrate a machine learning based approach to automatic classification of the three types of force curves and a low pass filter for noise removal, independent component analysis for dimensionality reduction and support vector machine for data classification are involved in this process. It is validated by the experiments that the three types of force curves recorded using AFM can be effectively and efficiently classified with the proposed approach.