Algorithmic approaches to high speed atomic force microscopy.

摘要

The atomic force microscope (AFM) has a unique set of capabilities for investigating biological systems, including sub-nanometer spatial resolution and the ability to image in liquid and to measure mechanical properties. Acquiring a high quality image, however, can take from minutes to hours. Despite this limited frame rate, researchers use the instrument to investigate dynamics via time-lapse imaging, driven by the need to understand biomolecular activities at the molecular level. Studies of processes such as DNA digestion with DNase, DNA-RNA polymerase binding and RNA transcription from DNA by RNA polymerase redefined the potential of AFM in biology. As a result of the need for better temporal resolution, advanced AFMs have been developed. The current state of the art in high-speed AFM (HS-AFM) for biological studies is an instrument developed by Toshio Ando at Kanazawa University in Japan. This instrument can achieve 12 frames/sec and has successfully visualized the motion of protein motors at the molecular level. This impressive instrument as well as other advanced AFMs, however, comes with tradeoffs that include a small scan size, limited imaging modes and very high cost. As a result, most AFM users still rely on standard commercial AFMs. The work in this thesis develops algorithmic approaches that can be implemented on existing instruments, from standard commercial systems to cutting edge HS-AFM units, to enhance their capabilities.;There are four primary contributions in this thesis. The first is an analysis of the signals available in an AFM with respect to the information they carry and their suitability for imaging at different scan speeds. The next two are algorithmic approaches to HS-AFM that take advantage of these signals in different ways. The first algorithm involves a new sample profile estimator that yields accurate topology at speeds beyond the bandwidth of the limiting actuator. The second involves more efficient sampling, using the data in real time to steer the tip. Both algorithms yield at least an order of magnitude improvement in imaging rate but with different tradeoffs. The first operates beyond the bandwidth of the controller managing the tip-sample interaction and therefore the applied force is not well-regulated. The second keeps this control intact but is effective only on a limited set of samples, namely biopolymers or other string-like samples. Experiments on calibration samples and lambda-DNA show that both of the algorithms improve the imaging rate by an order of magnitude. In the fourth contribution, extended applications of AFMs equipped with the algorithmic approaches are the tracking of a macromolecule moving along a string-like sample and a time optimal path for repetitive non-raster scans along string-like samples.