A machine-learning approach reportedly rescued “impossible” human drug transporter mutations on an in-silico basis, highlighting how computational tools can expand the set of viable transporter variants for drug discovery. The work focuses on the PepT1 transporter (SLC15A1), a key intestinal peptide gatekeeper for peptide-like drugs. By correcting or compensating for damaging mutations, the model suggests a path for exploring alternative transporter engineering strategies when clinical pharmacokinetics or uptake is impaired. The result also underscores the growing use of AI as a hypothesis generator for protein function, not just structure prediction. Translation will hinge on whether the rescued variants maintain function in biological systems at therapeutically relevant expression levels and whether they alter drug absorption predictably without introducing new safety liabilities.