High-content High-speed Chemical Imaging of Metabolic Reprogramming by Integration of Advanced Instrumentation and Data Science

项目总结:

Project Summary: Providing molecular fingerprint vibration information and high imaging speed, coherent Raman scattering microscopy, based on either coherent anti-Stokes Raman scattering (CARS) or stimulated Raman scattering (SRS), allows real-time vibrational imaging of living cells and/or tissues with sub-micron spatial resolution These instrumentation-based advances, however, do not fulfill all the desired parameters in hyperspectral imaging, including broad bandwidth, high signal to noise ratio (SNR) and high speed. In pushing these physical limits, it is common that one parameter is optimized at the price of sacrificing other advantages. The current proposal aims to break this conventional thinking of "no free lunch in optimization" through a synergistic integration of advanced instrumentation and data science. A multidisciplinary team with a strong track record of collaborations will pursue the proposed studies. Ji-Xin Cheng (PI) is a leading expert in the development and applications of SRS chemical imaging. Lei Tian (co-I) is a leading expert in computational microscopy and machine learning. Daniela Matei (co-I) is a leading expert in cancer research specialized in ovarian cancer. We aim to develop two complementary platforms that will allow high-speed, high-content, and high-sensitivity mapping of cell metabolism. The first platform is for samples without prior knowledge. We will build a polygon scanner to tune the delay between two chirped pulses on a 20-microsecond time scale. We will then deploy deep spatial-spectral learning to denoise the low-SNR hyperspectral measurements and extract salient information with much enhanced SNR. This integrated approach effectively bypasses the conventional tradeoff between acquisition speed and SNR and enables high-speed, high-throughput, hyperspectral SRS imaging using informative fingerprint Raman bands. The second platform is for samples with known target species. We will develop a sparsely sampled hyperspectral imaging strategy to increase the overall speed by one order of magnitude while maintaining the same SNR. We will develop a novel "recursive feature elimination" approach to determine the minimum number of essential frames. On the instrumentation side, a fast-tuning fiber laser will be deployed to acquire a sparsely sampled hyperspectral stack within one second for the study of living systems. As a focused application, we will apply the proposed platforms to systematically investigate metabolic reprogramming in ovarian cancers that are cisplatin resistant. Our focused application will unveil hidden signatures that are associated with drug resistance, which will open new opportunities for improved treatment of drug-resistant cancers.