Poor molecular selectivity and unacceptable rate of sensor-to-sensor variations have been stumbling blocks in translating chemical and biological sensors based of microelectromechanical systems into a practical reality. Traditionally, selectivity in chemical and biological sensing is achieved by using immobilized receptors or chemical interfaces on cantilever surfaces. While poor selectivity in small molecule detection using reversible receptors is due to the lack of uniqueness of receptor-analyte interaction, nonuniformity in the immobilized receptor graft density leads to unacceptable rates of reproducibility in these sensors. Engineering the receptor layers with vectorial connectivity along the length of the cantilevers shows increased reproducibility and much enhanced sensitivity in chemical and biological detection. Various strategies are currently used for overcoming the selectivity challenges. These approaches include separation prior to detection, array-based sensing, and incorporation of multi-modal, multi-physics approaches. These cantilevers, when fabricated as bimaterial cantilevers, can be used as highly sensitive thermal sensors for photothermal spectroscopy of adsorbed molecules. Cantilever-based photothermal spectroscopy overcomes many of the selectivity challenges encountered when using receptor-based approaches. Multi-modal, multi-physics data obtained by judicial integration of these approaches when analyzed using advanced machine learning techniques enhance the selectivity, sensitivity, and reliability of MEMS sensor arrays.