Inadequate bioavailability is one of the most critical reasons for the failure of oral drug development. However, the way that substructures affect bioavailability remains largely unknown. Serotonin transporter (SERT) inhibitors are first-line drugs for major depression disorder, and improving their bioavailability may be able to decrease side-effects by reducing daily dose. Thus, it is an excellent model to probe the relationship between substructures and bioavailability. Here, we proposed the concept of "nonbioavailable substructures", referring to substructures that are unfavorable to bioavailability. A machine learning model was developed to identify nonbioavailable substructures based on their molecular properties and shows the accuracy of 83.5%. A more potent SERT inhibitor DH4 was discovered with a bioavailability of 83.28% in rats by replacing the nonbioavailable substructure of approved drug vilazodone. DH4 exhibits promising anti-depression efficacy in animal experiments. The concept of nonbioavailable substructures may open up a new venue for the improvement of drug bioavailability.