Parkinson's disease (PD) represents one of the most frequent neurodegenerative disorders for which genetic diagnosis is still challenging due to the high genetic heterogeneity associated with the disease and to the difficulty in interpreting test results. We have recently reported the identification of rare new gene variants in PD patients that support polygenic contribution to the disease. Here we report the identification of novel candidate PD genes and an exploratory protocol for predictive analysis of PD risk. The study includes the whole exome data of 22 PD families, 300 unrelated familiar PD, 504 unrelated sporadic PD and 664 healthy subjects. Family-based approach identified rare and disrupting variants in 44 candidate PD genes co-inherited by affected relatives. The analysis of the entire cohort discovered a significant excess of rare and deleterious variants in PD patients compared to controls in 7 genes out of the 44 identified in the families. Five of these, known as ANKK1, ANKRD50, GRK5, PACSIN1 and VPS8, were novel candidate PD genes, expressed in human dopaminergic neurons, and involved in signal transduction pathways and in endocytic recycling. In these genes, we identified both rare probably damaging variants, altering protein structure and dynamics, as well as frequent variants associated with PD risk. Moreover, we demonstrated that the co-inheritance of multiple rare variants (≥ 2) in a panel of 37 PD genes selected in this study, may predict disease risk in about 26 % of patients, both familial and sporadic cases, with high specificity (> 92 %; p ≤0.00001). Furthermore, patients carrying multiple rare variants showed higher risk of manifesting dyskinesia induced by levodopa treatment (p = 0.004), severe cognitive impairment (p = 0.009) and an earlier age at onset of the disease (p = 0.01). Despite the still exploratory nature of the study, these data provide novel insights into the genetic of PD and may be relevant for its prediction, diagnosis and treatment.