Cellular heterogeneity facilitating specialization of cells is essential to control normal physiological processes as well as drive pathogenesis of many diseases. While it remains common to measure population-averaged properties in biological studies, the assumption that all cells are identical can lead to incorrect or at least imprecise assessments. To better understand heterogeneous characteristics and responses of individual cells, single-cell analytic methods should be applied to profile individual cellular properties. Using micro-fabrication technology, we have developed high-throughput microfluidic platforms, that enable tracking of thousands of single cells on-chip. Building on the cell isolation capability, we innovated a photomechanical actuation method to selectively detach and retrieve target cells for downstream analysis. Irradiation of a nanosecond-pulsed laser is utilized to generate shear force for detaching target cells cultured on carbon nanotube–polydimethylsiloxane (CNT-PDMS) composite. This non-destructive cell retrieval method can preserve cell viability, membrane proteins, and mRNA expression levels, so we have successfully performed single-cell transcriptome analysis as well as functional studies on the retrieved target cells. In addition to cell tracking and retrieval, we applied machine learning techniques to correlate cellular morphological features with cellular functions including migration, cancer drug response, and tumor initiation. Machine made significantly better predictions than experienced researchers, and we found novel morphological features facilitating breakthroughs in mechanistic understanding. The integrated single-cell manipulation and analysis augmented with machine learning will change how we understand cell biology and ultimately improve how we treat diseases.