Computational modelling of modern cancer immunotherapy.

Modern cancer immunotherapy has revolutionised oncology and carries the potential to radically change the approach to cancer treatment. However, numerous questions remain to be answered to understand immunotherapy response better and further improve the benefit for future cancer patients. Computational models are promising tools that can contribute to accelerated immunotherapy research by providing new clues and hypotheses that could be tested in future trials, based on preceding simulations in addition to the empirical rationale. In this topical review, we briefly summarise the history of cancer immunotherapy, including computational modelling of traditional cancer immunotherapy, and comprehensively review computational models of modern cancer immunotherapy, such as immune checkpoint inhibitors (as monotherapy and combination treatment), co-stimulatory agonistic antibodies, bispecific antibodies, and chimeric antigen receptor T cells. The modelling approaches are classified into one of the following categories: data-driven top-down vs mechanistic bottom-up, simplistic vs detailed, continuous vs discrete, and hybrid. Several common modelling approaches are summarised, such as pharmacokinetic/pharmacodynamic models, Lotka-Volterra models, evolutionary game theory models, quantitative systems pharmacology models, spatio-temporal models, agent-based models, and logic-based models. Pros and cons of each modelling approach are critically discussed, particularly with the focus on the potential for successful translation into immuno-oncology research and routine clinical practice. Specific attention is paid to calibration and validation of each model, which is a necessary prerequisite for any successful model, and at the same time, one of the main obstacles. Lastly, we provide guidelines and suggestions for the future development of the field.

[18F]FDG PET immunotherapy radiomics signature (iRADIOMICS) predicts response of non-small-cell lung cancer patients treated with pembrolizumab.

Background Immune checkpoint inhibitors have changed the paradigm of cancer treatment; however, non-invasive biomarkers of response are still needed to identify candidates for non-responders. We aimed to investigate whether immunotherapy [18F]FDG PET radiomics signature (iRADIOMICS) predicts response of metastatic non-small-cell lung cancer (NSCLC) patients to pembrolizumab better than the current clinical standards. Patients and methods Thirty patients receiving pembrolizumab were scanned with [18F]FDG PET/CT at baseline, month 1 and 4. Associations of six robust primary tumour radiomics features with overall survival were analysed with Mann-Whitney U-test (MWU), Cox proportional hazards regression analysis, and ROC curve analysis. iRADIOMICS was constructed using univariate and multivariate logistic models of the most promising feature(s). Its predictive power was compared to PD-L1 tumour proportion score (TPS) and iRECIST using ROC curve analysis. Prediction accuracies were assessed with 5-fold cross validation. Results The most predictive were baseline radiomics features, e.g. Small Run Emphasis (MWU, p = 0.001; hazard ratio = 0.46, p = 0.007; AUC = 0.85 (95% CI 0.69-1.00)). Multivariate iRADIOMICS was found superior to the current standards in terms of predictive power and timewise with the following AUC (95% CI) and accuracy (standard deviation): iRADIOMICS (baseline), 0.90 (0.78-1.00), 78% (18%); PD-L1 TPS (baseline), 0.60 (0.37-0.83), 53% (18%); iRECIST (month 1), 0.79 (0.62-0.95), 76% (16%); iRECIST (month 4), 0.86 (0.72-1.00), 76% (17%). Conclusions Multivariate iRADIOMICS was identified as a promising imaging biomarker, which could improve management of metastatic NSCLC patients treated with pembrolizumab. The predicted non-responders could be offered other treatment options to improve their overall survival.

Computational modelling of resistance and associated treatment response heterogeneity in metastatic cancers.

Metastatic cancer patients invariably develop treatment resistance. Different levels of resistance lead to observed heterogeneity in treatment response. The main goal was to evaluate treatment response heterogeneity with a computation model simulating the dynamics of drug-sensitive and drug-resistant cells. Model parameters included proliferation, drug-induced death, transition and proportion of intrinsically resistant cells. The model was benchmarked with imaging metrics extracted from 39 metastatic prostate cancer patients who had 18F-NaF-PET/CT scans performed at baseline and at three cycles into chemotherapy or hormonal therapy. Two initial model assumptions were evaluated: considering only inter-patient heterogeneity and both inter-patient and intra-patient heterogeneity in the proportion of intrinsically resistant cells. The correlation between the median proportion of intrinsically resistant cells and baseline patient-level imaging metrics was assessed with Spearman's rank correlation coefficient. The impact of model parameters on simulated treatment response was evaluated with a sensitivity study. Treatment response after periods of six, nine, and 12 months was predicted with the model. The median predicted range of response for patients treated with both therapies was compared with a Wilcoxon rank sum test. For each patient, the time was calculated when the proportion of disease with a non-favourable response outperformed a favourable response. By taking into account inter-patient and intra-patient heterogeneity in the proportion of intrinsically resistant cells, the model performed significantly better ([Formula: see text]) than by taking into account only inter-patient heterogeneity ([Formula: see text]). The median proportion of intrinsically resistant cells showed a moderate correlation (ρ = 0.55) with mean patient-level uptake, and a low correlation (ρ = 0.36) with the dispersion of mean metastasis-level uptake in a patient. The sensitivity study showed a strong impact of the proportion of intrinsically resistant cells on model behaviour after three cycles of therapy. The difference in the median range of response (MRR) was not significant between cohorts at any time point (p  > 0.15). The median time when the proportion of disease with a non-favourable response outperformed the favourable response was eight months, for both cohorts. The model provides an insight into inter-patient and intra-patient heterogeneity in the evolution of treatment resistance.

Predicting tumour response to anti-PD-1 immunotherapy with computational modelling.

Cancer immunotherapy is a rapidly developing field, with numerous drugs and therapy combinations waiting to be tested in pre-clinical and clinical settings. However, the costly and time-consuming trial-and-error approach to development of new treatment paradigms creates a research bottleneck, motivating the development of complementary approaches. Computational modelling is a compelling candidate for this task, however, difficulties associated with the validation of such models limit their use in pre-clinical and clinical settings. Here we propose a bottom-up deterministic computational model to simulate tumour response to treatment with anti-programmed-death-1 antibodies (anti-PD-1). The model was built with validation in mind, and so contains minimum number of parameters, and only four free parameters. Moreover, all model parameters can be measured experimentally. Free parameters were tuned by fitting the model to experimental data from the literature, using B16-F10 murine melanoma implanted into wild type (C57BL/6), as well as into immunodeficient (NSG) mice strains, and treated with anti-PD-1 antibodies. The model's predictive ability was verified on two independent datasets from literature with different but well-known inputs. To identify possible biomarkers of response to anti-PD-1 immunotherapy, sensitivity study of key model parameters was performed. Good agreement between the simulated tumour growth curves and the experimental data was achieved, with mean relative deviations in the range of 13%-20%. Our sensitivity study demonstrated that major histocompatibility complex (MHC) class I expression was the only parameter able to clearly discriminate responders from non-responders to anti-PD-1 therapy. Additionally, the results of sensitivity studies suggest that MHC class I expression might affect the predictive ability of other biomarkers that are currently used in the clinics, such as PD-1 ligand (PD-L1) expression. Interestingly, our model predicts the best response to anti-PD-1 therapy for subjects with moderate PD-L1 values. Such computational models show promise to support, guide and accelerate future immunotherapy research.