Optical in-phase and quadrature (IQ) constellations enclose valuable information regarding the optical elements traversed by the optical signal. Such information can be extracted and exploited by algorithms and models within an optical layer digital twin. In this paper, we investigate the feasibility of extracting information from IQ constellations and its use for both accurate quality of transmission (QoT) estimation and efficient failure management within. First, we observe the correlations between the measured QoT of lightpaths and the value of specific features extracted from IQ constellation samples collected from the optical receiver and design deep neural network (DNN) models for QoT estimation. Next, specific DNN models and algorithms that exploit IQ constellation features are proposed for soft-failure detection, identification, and severity estimation. Results from both simulation and experiments show noticeable accuracy on the estimation of QoT and on the prediction of failures affecting the transmitter, optical filters, and amplifiers.