Benchmarking ReleaseKinetix: Numerical Equivalence of a Web-Based Platform for Dissolution Analytics and Regulatory Decision Support

Before innovative computational instruments are incorporated into regulatory research and scientific practice, a thorough presentation of their computational reproducibility is fundamental. This study thoroughly benchmarks ReleaseKinetix. ReleaseKinetix is an independent, browserbased platform. It is used for comparison of in vitro dissolution profiles and release-kinetic modeling. The study benchmarks ReleaseKinetix (for comparing in vitro dissolution profiles and modeling release kinetics) against two independently written nonlinear least-squares reference engines. These are Python/scipy and R/minpack.lm. The study also benchmarks ReleaseKinetix against the third-party R package bootf2, published DDSolver Excel add-in values. The study also benchmarks ReleaseKinetix against data from the published literature. Seven datasets were used. Six came from the literature. One was synthetic with high variability. Agreement was judged against a predefined absolute tolerance of 0.01, corresponding to the two decimal places with which the similarity factor f2 is commonly reported. The 288 compared values covered similarity and difference factors, dissolution efficiency, mean dissolution time, and the parameters and goodness-of-fit statistics of six kinetic models. Of these, 272 were reproduced exactly (with a difference of less than one millionth), and the remaining 16 (all nonlinear model parameters) differed only at the level of solver tolerance, with none of them exceeding 0.01. The largest difference (0.00026) was smaller than the difference between the two reference engines (0.00047). The platform and the reference engines selected the same candidate kinetic models using the Akaike information criterion and reached the same conclusion based on similarity for every dataset. An apparent discrepancy in the Akaike values between the software tools was caused by a constant offset resulting from two different formula conventions, not a calculation error. The benchmark also revealed two errors in the platform, both of them corrected before the reported results. First, small decreases of the mean profile on the dissolution plateau were ignored in the mean dissolution time calculation, which resulted in an overestimation of 0.022 min. Second, the European Medicines Agency's 85% truncation rule was triggered by the reference product alone, whereas the guideline requires truncation as soon as either product exceeds 85%. This misreading was also observed in the in-house reference engines and was only detected through comparison with bootf2. No similarity conclusions changed after correction. The expected f2 had similar bootstrap confidence intervals across the platform and the reference engines. For the published data, bootf2 had similar confidence intervals within random sampling variation. However, near the similarity limit of 50, the lower confidence bound depended on the random seed. ReleaseKinetix reproduces standard dissolution calculations as precisely as independent implementations. This verified computational core forms the basis of the platform’s rule-based regulatory decision support layer, the interpretive reliability of which remains to be established in future studies.