Personalized oncology with artificial intelligence: The case of temozolomide.

PURPOSE: Using artificial intelligence techniques, we compute optimal personalized protocols for temozolomide administration in a population of patients with variability. METHODS: Our optimizations are based on a Pharmacokinetics/Pharmacodynamics (PK/PD) model with population variability for temozolomide, inspired by Faivre et al. [10] and Panetta et al. [25,26]. The patient pharmacokinetic parameters can only be partially observed at admission and are progressively learned by Bayesian inference during treatment. For every patient, we seek to minimize tumor size while avoiding severe toxicity, i.e. maintaining an acceptable toxicity level. The optimization algorithm we rely on borrows from the field of artificial intelligence. RESULTS: Optimal personalized protocols (OPP) achieve a sizable decrease in tumor size at the population level but also patient-wise. The tumor size is on average 67.2 g lighter than with the standard maximum-tolerated dose protocol (MTD) after 336 days (12 MTD cycles). The corresponding 90% confidence interval for average tumor size reduction amounts to 58.6-82.7 g. When treated with OPP, less patients experience severe toxicity in comparison to MTD. MAJOR FINDINGS: We quantify in-silico the benefits offered by personalized oncology in the case of temozolomide administration. To do so, we compute optimal personalized protocols for a population of heterogeneous patients using artificial intelligence techniques. At each treatment day, the protocol is updated by taking into account the feedback obtained from patient's reaction to the drug administration. Personalized protocols greatly differ from each other, and from the standard MTD protocol. Benefits of personalization are very sizable: tumor sizes are much smaller on average and also patient-wise, while severe toxicity is made less frequent.

Optimizing treatment combination for lymphoma using an optimization heuristic.

BACKGROUND: The standard treatment for high-grade non-Hodgkin lymphoma involves the combination of chemotherapy and immunotherapy. We characterize in-silico the optimal combination protocol that maximizes the overall survival probability. We rely on a pharmacokinetics/pharmacodynamics (PK/PD) model that describes the joint evolution of tumor and effector cells, as well as the effects of both chemotherapy and immunotherapy. The toxicity is taken into account through ad-hoc constraints. We develop an optimization algorithm that belongs to the class of Monte-Carlo tree search algorithms. Our simulations rely on an in-silico population of heterogeneous patients differing with respect to their PK/PD parameters. The optimization objective consists in characterizing the combination protocol that maximizes the overall survival probability of the patient population under consideration. RESULTS: We compare using in-silico experiments our results to standard protocols and observe a gain in overall survival probabilities that vary from 4 to 9 percentage points. The gains increase with the complexity of the potential protocol. Gains are larger in presence of a higher number of injections or of an actual combination with immunotherapy. CONCLUSIONS: In in-silico experiments, optimal protocols achieve significant gains over standard protocols when considering overall survival probabilities. Our optimization algorithm enables us to efficiently tackle this numerical problem with a large dimensionality. The in-vivo implications of our in-silico results remain to be explored.

Optimizing immune cell therapies with artificial intelligence.

PURPOSE: We determine an optimal injection pattern for anti-vascular endothelial growth factor (VEGF) and for the combination of anti-VEGF and unlicensed dendritic cells. METHODS: We rely on the mathematical model of Soto-Ortiz and Finley (2016) for the interactions between the tumor growth, angiogenesis and immune system reactions. Our optimization algorithm belongs to the class of Monte-Carlo tree search algorithms. The objective consists in finding the minimal total drug doses for which an injection pattern yields tumor eradication. RESULTS: Our results are twofold. First, optimized injection protocols enable to significantly reduce the total drug dose for tumor elimination. For instance, for an early diagnosis date, a total dose equal to 58% of the standard anti-VEGF dose enables to eliminate the tumor. In the case of drug combination, associating 25% of the total standard anti-VEGF dose to 10% of the dendritic cell total standard dose eradicates tumor. Our second result is that administering a dose equal to the maximal standard dose allows for later diagnosis date compared to standard protocol. For instance, in the case of anti-VEGF injection, the optimal protocol postpones the maximal diagnosis date by more than one month. CONCLUSIONS: Overall, our optimization based on artificial intelligence delivers significant gains in total drug administration or in the length of the therapeutic window. Our method is flexible and could be adapted to other drug combinations.

Optimal dynamic regimens with artificial intelligence: The case of temozolomide.

We determine an optimal protocol for temozolomide using population variability and dynamic optimization techniques inspired by artificial intelligence. We use a Pharmacokinetics/Pharmacodynamics (PK/PD) model based on Faivre and coauthors (Faivre, et al., 2013) for the pharmacokinetics of temozolomide, as well as the pharmacodynamics of its efficacy. For toxicity, which is measured by the nadir of the normalized absolute neutrophil count, we formalize the myelosuppression effect of temozolomide with the physiological model of Panetta and coauthors (Panetta, et al., 2003). We apply the model to a population with variability as given in Panetta and coauthors (Panetta, et al., 2003). Our optimization algorithm is a variant in the class of Monte-Carlo tree search algorithms. We do not impose periodicity constraint on our solution. We set the objective of tumor size minimization while not allowing more severe toxicity levels than the standard Maximum Tolerated Dose (MTD) regimen. The protocol we propose achieves higher efficacy in the sense that -compared to the usual MTD regimen- it divides the tumor size by approximately 7.66 after 336 days -the 95% confidence interval being [7.36-7.97]. The toxicity is similar to MTD. Overall, our protocol, obtained with a very flexible method, gives significant results for the present case of temozolomide and calls for further research mixing operational research or artificial intelligence and clinical research in oncology.

Administration of temozolomide: Comparison of conventional and metronomic chemotherapy regimens.

PURPOSE: We compare the Maximum Tolerated Dose (MTD) and Metronomic Chemotherapy (MC) protocols for temozolomide administration. We develop an innovative methodology for characterizing optimal chemotherapy regimens. METHODS: We use a PK/PD model based on Faivre et al. (2013) for the pharmacokinetics of temozolomide, as well as the pharmacodynamics of its efficacy. For toxicity, which is measured by the nadir of the normalized absolute neutrophil count, we formalize the myelosuppression effect of temozolomide with the physiological model of Panetta et al. (2003b). We introduce a multi-criteria tool for comparing protocols along their efficacy and toxicity dimensions. RESULTS: We show that the toxicity of the MC regimen proposed by Faivre et al. (2013) can greatly be reduced without affecting its efficacy, while the standard MTD protocol efficacy cannot be improved without impairing its toxicity. We also show that for any acceptable toxicity level, the optimal protocol remains closely related to standard MTD. CONCLUSIONS: Overall, our new method enables a rich comparison between protocols along multiple dimensions. We can rank protocols for temozolomide administration. It is a first step toward building optimal individual protocols.