A team of computational biologists and oncologists has built one of the most ambitious digital disease models yet attempted in lung cancer research: a quantitative systems pharmacology model that simulates how non-small cell lung cancer (NSCLC) tumors driven by all three functional classes of BRAF mutations respond to targeted drugs. The work, published in iScience, addresses a long-standing gap in precision oncology. While patients whose tumors carry the common BRAF V600E mutation have an approved targeted therapy, roughly half of BRAF-mutant NSCLC patients harbor rarer non-V600 variants for which no effective targeted treatment exists. By encoding the distinct biology of class I, II, and III BRAF mutations into a single mechanistic framework, the researchers created a virtual testing ground where thousands of simulated patients could be trialed against seven different kinase inhibitors, revealing combination strategies that were then confirmed in mice carrying real patient tumor tissue.

The biological problem the model tackles is rooted in the peculiar behavior of the BRAF protein, a kinase sitting near the top of the MAPK signaling cascade that drives cell proliferation. Class I mutations, essentially the V600E variant, transform BRAF into a constitutively active monomer that fires independently of RAS and of dimerization with other RAF family members. Class II mutations also signal without RAS, but they do so as active dimers with moderate to strong kinase output. Class III mutations are the strangest of all: they have little or no kinase activity themselves and instead promote cancer by binding RAS more tightly and forming heterodimers with wild-type CRAF, effectively amplifying upstream signals. Class III mutants frequently coexist with RAS mutations or loss of the tumor suppressor NF1, further rewiring the pathway. These mechanistic differences explain why first-generation RAF inhibitors, which work brilliantly against V600E monomers, largely fail in class II and III tumors.

The consequences for patients are stark. The only approved targeted regimen for BRAF-mutant NSCLC, the combination of the RAF inhibitor dabrafenib with the MEK inhibitor trametinib, is restricted to class I V600E disease. For everyone else, options are limited to chemotherapy or immunotherapy, and clinical data on non-V600 subgroups are scarce because these mutations are individually rare and scattered across trials. Compounding the problem, class II and III tumors deploy compensatory survival mechanisms, including reactivation of RAS-MAPK signaling, loss of NF1, and bypass signaling through receptor tyrosine kinases such as EGFR, FGFR1, and IGF1R, which sustain MAPK output and can co-activate the PI3K-AKT pathway even under drug pressure. Clinical trials for each rare subtype would take years and enormous expense, which is precisely where the modeling approach promises to change the calculus of drug development.

Quantitative systems pharmacology, or QSP, sits at the intersection of pharmacology, cell biology, and computational engineering. Rather than fitting simple statistical curves to response data, QSP models represent the underlying biology as networks of ordinary differential equations describing receptor binding, phosphorylation, feedback inhibition, drug-target interactions, and pharmacokinetics. The new model comprises roughly 79 molecular species and 70 equations, built on mass action kinetics and Hill-type reaction principles, with two-compartment pharmacokinetic modules for each drug. It captures the EGFR, FGFR1, and IGF1R receptor pathways, including the ligand-independent self-activation of FGFR1, and encodes the distinct dimerization and feedback behaviors of each BRAF mutation class. Seven inhibitors are represented: the MEK inhibitors trametinib and cobimetinib, the RAF inhibitors vemurafenib, dabrafenib, and encorafenib, and the ERK inhibitors ulixertinib and ravoxertinib.

Calibrating such a model demands data at every scale, and the team assembled an unusually rich dataset. At the cellular level, they used NSCLC cell lines representing each mutation class: HCC364 for class I, NCI-H1395 and NCI-H1755 for class II, and NCI-H1666 and CAL-12T for class III. Around 121 experimental datasets, containing more than 605 data points, quantified how cells respond to ligand stimulation with EGF, FGF2, and IGF1, and how phosphorylation of EGFR, GAB1, SHP2, RAS, MEK, ERK, AKT, and mTOR changes over time and dose under drug treatment. The model reproduced these dynamics quantitatively, then passed a critical test: when asked to predict the effects of drug combinations it had never seen, such as vemurafenib plus cobimetinib or vemurafenib plus encorafenib, its simulated viability curves matched the experimental measurements closely.

The model then climbed to the level of whole animals. Assuming that intracellular signaling and drug-potency parameters carry over from the cell studies, the researchers added mouse pharmacokinetics and calibrated tumor growth rates against untreated control xenografts. The resulting framework captured dose- and time-dependent tumor growth inhibition across cell-derived xenografts from lung, thyroid, colorectal, and melanoma origins, and across patient-derived xenografts for class II and III mutations. Notably, without any parameter adjustment, it also predicted the synergy of trametinib plus dabrafenib in a melanoma xenograft model. Goodness-of-fit statistics were strong: most R-squared values fell between 0.8 and 1, and mean absolute percentage error stayed below 20 percent, benchmarks the authors describe as qualifying the model for translational use.

The most consequential step was the construction of virtual clinical trials. Using Monte Carlo sampling with log-normal distributions, the team generated three populations of 5,000 virtual patients each, one per mutation class, randomizing receptor levels, tumor burden, growth rates, and BRAF-driven pathway activity within biologically plausible ranges. Virtual patients with implausible growth kinetics were excluded, and the remaining populations were weighted so their predicted response distributions matched real clinical data. The approach was first validated against known results: simulated objective response rates for vemurafenib, dabrafenib, and the vemurafenib-cobimetinib combination in class I patients accurately reproduced the outcomes of actual trials, including NCT01524978 and NCT01336634. Simulations also matched observed responses to vemurafenib in class II patients from the AcSé trial and to trametinib in class III patients from the NCI-MATCH trial.

With the platform validated, the researchers ran experiments no real trial could easily perform. Dose-optimization simulations suggested that the approved standard dosing of trametinib plus dabrafenib is near-optimal for class I patients, with escalation adding little benefit. For class II patients, however, a higher trametinib dose of 2.5 milligrams daily, alone or with dabrafenib, predicted better response rates than the standard 2-milligram regimen. In class III virtual patients, response rates trended upward for both vemurafenib and cobimetinib beyond their approved doses, hinting that tolerability permitting, higher exposure might help this neglected subgroup. A global sensitivity analysis using the partial rank correlation coefficient algorithm identified ERK, MEK, AKT, and mTOR signaling parameters as the dominant drivers of tumor response, pointing the way to rational combinations.

The headline prediction concerned class II and III disease: combining a MEK inhibitor with an ERK inhibitor, rather than the conventional RAF-MEK pairing, should produce the greatest tumor shrinkage in these subtypes. To test this, the team turned to MiniPDX technology, in which fresh tumor cells from actual patients are seeded into capsules and implanted in mice for rapid drug-sensitivity testing. Tumor samples came from four NSCLC patients, two with confirmed class II and two with class III BRAF mutations. Seven days of treatment with trametinib plus ravoxertinib significantly suppressed tumor growth in both class II and class III models without affecting animal body weights, and ravoxertinib alone showed significant activity in the class III models, exactly as the simulations had forecast. It is a rare and satisfying outcome in computational oncology: a model prediction, generated in silico, confirmed in vivo against real patient tissue within the same study.

The model also probed the receptor-level resistance mechanisms that sabotage targeted therapy in non-V600 tumors. Simulating partial blockade of EGFR, IGF1R, and FGFR signaling revealed a class-dependent hierarchy. In class I tumors, dual EGFR and IGF1R inhibition meaningfully reduced tumor volume beyond EGFR blockade alone. In class II tumors, any double blockade was insufficient, but simultaneous inhibition of all three receptors produced a dramatic reduction in tumor volume. Class III tumors proved most stubborn: double inhibition had no effect, and even triple inhibition yielded only modest regression, underscoring how deeply redundant the compensatory wiring is in these cancers. The authors caution that the safety of such triple combinations is untested and likely to involve additive toxicities such as rash, diarrhea, and hyperglycemia, and that their framework omits plasma protein binding, inter-patient pharmacokinetic variability, and normal tissue physiology. Still, as the first multiscale QSP model spanning all three BRAF mutation classes, validated from phosphoprotein signals to virtual patients to patient-derived xenografts, the platform offers a template for model-informed drug development in rare oncology subgroups, an approach whose relevance is only growing as regulators push for smarter dose optimization in cancer trials.

Subject of Research: Quantitative systems pharmacology modeling of class I-III BRAF-mutant non-small cell lung cancer for subgroup treatment analysis and combination therapy design

Article Title: Quantitative systems pharmacology modeling of class I–III BRAF-mutant NSCLC enables subgroup treatment analysis and combination therapy design

Article References: Hu, J., Rao, Q., Li, Z., Li, G., Bi, Z., Wen, D., Zhao, C., & Zhou, G. (2026). Quantitative systems pharmacology modeling of class I–III BRAF-mutant NSCLC enables subgroup treatment analysis and combination therapy design. iScience, 29(11), Article 117854. https://doi.org/10.1016/j.isci.2026.117854

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117854

Keywords: BRAF mutations, NSCLC, quantitative systems pharmacology, virtual clinical trials, MEK inhibitors, ERK inhibitors, RAF inhibitors, MAPK pathway, combination therapy, MiniPDX, precision oncology, drug resistance