PKpredict: A Computational Tool to Predict Tuberculosis Drug Penetration
At CIBB 2025, held at Politecnico di Milano from September 10 to 12, Roberto Visintainer, researcher at COSBI, presented his work entitled “PKpredict: A Machine Learning Framework to Predict Anti-Tuberculosis Drug Penetration into Lesion and Caseum Compartments,” developed in collaboration with the Gates Medical Research Institute.
Tuberculosis (TB) continues to be one of the deadliest infectious diseases worldwide, responsible for more than one million deaths each year. Current therapies require prolonged multi-drug regimens, and the rise of multi-drug resistance highlights the urgent need for shorter and more effective treatments. A key determinant of therapeutic success is the ability of antibiotics to penetrate pulmonary TB lesions and caseum, where bacteria can persist and evade treatment.
Directly measuring drug penetration in animal models such as rabbits is costly and resource-intensive. To address this, Roberto developed PKpredict, a computational tool that integrates machine learning (ML) algorithms with a minimal physiologically based pharmacokinetic (mPBPK) model. By linking compound physicochemical properties to their distribution at TB sites of action, PKpredict enables early predictions of drug penetration.
The tool was trained on pharmacokinetic data from 30 anti-TB drugs tested in rabbit lesion models. This approach provides a valuable in silico pipeline for drug discovery programs, allowing researchers to prioritize promising compounds and streamline clinical development. Looking forward, PKpredict holds potential for translational applications, by bridging preclinical findings and human data to accelerate the design of effective therapies against TB.
Congratulation to Roberto for taking COSBI’s research to the CIBB 2025 conference.
