arXiv subjects

Sébastien Benzekry

Publications and source records attributed to Sébastien Benzekry.

10 recordsLinked to original sources

arXiv2026

APOD: reasoning-guided agentic population ordinary differential equation discovery for pharmacological digital twins

Establishing ordinary differential equations (ODEs) describing population data is a fundamental part of mathematical modeling in pharmacology, crucial to developing digital twins. However, doing so from sparse, noisy data is a slow, expert-driven task. Existing automated methods either search a restricted model space or ignore population inter-individual variability. Here we introduce APOD (Agentic Population ODE Discovery), a language-model agent that iteratively reasons over biological knowledge and fit diagnostics in an open-ended search space to discover a population digital twin (PDT), i.e., a shared ODE system with between-subject variability. On synthetic pharmacokinetic and tumor-dynamics benchmarks, APOD recovered ground-truth structures in 94-100\% of runs, 12-fold faster in median than an established library-based search. On real cohorts it converged to valid structures, and proposed a PDT of radioligand-therapy-induced platelet dynamics that predicts thrombocytopenia from first-cycle data and simulates alternative dosing schedules that lower the predicted risk of toxicity.

q-bio.QM↗

arXiv2026

ROOFS: RObust biOmarker Feature Selection

Feature selection (FS) is essential for biomarker discovery and clinical predictive modeling. Over the past decades, methodological literature on FS has become rich and mature, offering a wide spectrum of algorithmic approaches. However, much of this methodological progress has not fully translated into applied biomedical research. Moreover, challenges inherent in biomedical data, such as high-dimensional feature space, low sample size, multicollinearity, and missing values, make FS non-trivial. To help bridge this gap between methodological development and practical application, we propose ROOFS (RObust biOmarker Feature Selection), a Python package available at https://gitlab.inria.fr/compo/roofs, designed to help researchers in the choice of FS method adapted to their problem. ROOFS benchmarks multiple FS methods on the user's data and generates reports summarizing a comprehensive set of evaluation metrics, including downstream predictive performance estimated using optimism correction, stability, robustness of individual features, and true positive and false positive rates assessed on semi-synthetic data with a simulated outcome. We demonstrate the utility of ROOFS on data from the PIONeeR clinical trial, aimed at identifying predictors of resistance to anti-PD-(L)1 immunotherapy in lung cancer. Of the 34 FS methods gathered in ROOFS, we evaluated 23 in combination with 11 classifiers (253 models) and identified a filter based on the union of Benjamini-Hochberg false discovery rate-adjusted p-values from t-test and logistic regression as the optimal approach, outperforming other methods including widely used LASSO. We conclude that comprehensive benchmarking with ROOFS has the potential to improve the reproducibility of FS discoveries and increase the translational value of clinical models.

stat.ML↗

arXiv2026

Mechanistic Learning for Survival Prediction in NSCLC Using Routine Blood Biomarkers and Tumor Kinetics

Background Predicting overall survival (OS) in non-small cell lung cancer (NSCLC) is essential for clinical decision-making and drug development. While tumor and blood test markers kinetics are intrinsically linked, their joint dynamics and relationship to OS remain unknown. Methods We developed a mechanistic model capturing the interplay between tumor (T) burden and three key blood markers kinetics: albumin (A), lactate dehydrogenase (L), and neutrophils (N), through coupled differential equations (termed TALN-k). This model was enhanced with a machine learning framework (TALN-kML) for OS prediction. The model was trained and validated on clinical trial data from NSCLC patients treated with atezolizumab in monotherapy (N = 862 patients) or combination therapy (N = 1,115). Model parameters were estimated using nonlinear mixed-effects modelling, and survival predictions were assessed using individual and trial level metrics. Results TALN-k successfully described individual and population-level marker kinetics, revealing complex interactions between tumor and blood markers, and improving corrected BIC and log-likelihood metrics by a significant margin of previous empirical state-of-the-art models. Feature selection methods also highlighted valuable predictive parameters, indicatives of good or poor prognosis. The TALN-kML model outperformed empirical, uncoupled models, achieving improved C-index (0.74 $\pm$ 0.02 vs 0.72 $\pm$ 0.03), 12-months AUC (0.83 $\pm$ 0.004 vs 0.79 $\pm$ 0.05), and accuracy (0.77 $\pm$ 0.03 vs 0.76 $\pm$ 0.05) in OS prediction. Conclusion Our mechanistic learning approach allows for an interpretable model, which improves on longitudinal data description and on survival prediction in NSCLC by jointly integrating tumor and blood markers kinetics. This methodology offers a promising avenue for both personalized treatment strategies and drug development optimization.

q-bio.QM↗

arXiv2025

The SChISM study: Cell-free DNA size profiles as predictors of progression in advanced carcinoma treated with immune-checkpoint inhibitors

Background: Many advanced cancer patients experience progression under immune-checkpoint inhibitors (ICIs). Circulating cell-free DNA (cfDNA) size profiles offer a promising noninvasive multi-cancer approach to monitor and predict immunotherapy response. Methods: In the SChISM (Size CfDNA Immunotherapy Signature Monitoring) study (NCT05083494), pre-treatment plasmatic cfDNA size profiles from 126 ICI-treated advanced carcinomas were quantified using the BIABooster device. Fragmentomederived variables (concentration, peaks' position, and fragment size ranges) at baseline were analyzed for associations with early progression (EP, progression at first imaging) and progression-free survival (PFS), using logistic and Cox regression models. Bootstrap analysis validated robustness. Additional analyses were performed in homogeneous subpopulations: first-line lung cancer patients (n = 60) and head-andneck patients treated with Nivolumab (n = 25). Results: Higher cfDNA concentration and high quantities of short fragments (111-240 base pairs (bp)) were associated with poor response, unlike long fragments (> 300 bp). The proportion of fragments longer than 1650 bp demonstrated highest discriminatory power (AUC = 0.73, C-index = 0.69). It was significantly associated with non-EP (odds ratio = 0.39 [95% CI: 0.25-0.62]) and longer PFS (hazard ratio: 0.54 [95% CI: 0.42-0.68]). These associations remained significant when adjusted for confounders (age, sex, Eastern Cooperative Oncology Group performance status, tumor type, and neutrophil-to-lymphocyte ratio) and across both subpopulations. Bootstrap analysis confirmed robustness with mean accuracy of 70.1 $\pm$ 4.17% and positive predictive value of 55.6 $\pm$ 7.37%, in test sets. Conclusion: cfDNA size profiles significantly predicted ICI response and anticipate relapse, outperforming the routinely used marker programmed death-ligand 1 immunohistochemistry and reflecting enhanced immune system activation. Trial registration: (NCT05083494), date of registration: 2021-10-19.

q-bio.QM↗

arXiv2024

Comprehensive Joint Modeling of First-Line Therapeutics in Non-Small Cell Lung Cancer

First-line antiproliferatives for non-small cell lung cancer (NSCLC) have a relatively high failure rate due to high intrinsic resistance rates and acquired resistance rates to therapy. 57% patients are diagnosed in late-stage disease due to the tendency of early-stage NSCLC to be asymptomatic. For patients first diagnosed with metastatic disease the 5-year survival rate is approximately 5%. To help accelerate the development of novel therapeutics and computer-based tools for optimizing individual therapy, we have collated data from 11 different clinical trials in NSCLC and developed a semi-mechanistic, clinical model of NSCLC growth and pharmacodynamics relative to the various therapeutics represented in the study. In this study, we have produced extremely precise estimates of clinical parameters fundamental to cancer modeling such as the rate of acquired resistance to various pharmaceuticals, the relationship between drug concentration and rate of cancer cell death, as well as the fine temporal dynamics of anti-VEGF therapy. In the simulation sets documented in this study, we have used the model to make meaningful descriptions of efficacy gain in making bevacizumab-antiproliferative combination therapy sequential, over a series of days, rather than concurrent.

q-bio.CB↗

arXiv2023

Une comparaison des algorithmes d'apprentissage pour la survie avec données manquantes

Survival analysis is an essential tool for the study of health data. An inherent component of such data is the presence of missing values. In recent years, researchers proposed new learning algorithms for survival tasks based on neural networks. Here, we studied the predictive performance of such algorithms coupled with different methods for handling missing values on simulated data that reflect a realistic situation, i.e., when individuals belong to unobserved clusters. We investigated different patterns of missing data. The results show that, without further feature engineering, no single imputation method is better than the others in all cases. The proposed methodology can be used to compare other missing data patterns and/or survival models. The Python code is accessible via the package survivalsim. -- L'analyse de survie est un outil essentiel pour l'étude des données de santé. Une composante inhérente à ces données est la présence de valeurs manquantes. Ces dernières années, de nouveaux algorithmes d'apprentissage pour la survie, basés sur les réseaux de neurones, ont été conçus. L'objectif de ce travail est d'étudier la performance en prédiction de ces algorithmes couplés à différentes méthodes pour gérer les valeurs manquantes, sur des données simulées qui reflètent une situation rencontrée en pratique, c'est-à dire lorsque les individus peuvent être groupés selon leurs covariables. Différents schémas de données manquantes sont étudiés. Les résultats montrent que, sans l'ajout de variables supplémentaires, aucune méthode d'imputation n'est meilleure que les autres dans tous les cas. La méthodologie proposée peut être utilisée pour comparer d'autres modèles de survie. Le code en Python est accessible via le package survivalsim.

stat.ML↗

arXiv2021

Asymptotic analysis of a biphase tumor fluid flow. The weak coupling case

The aim of this paper is to investigate the asymptotic behavior of a biphase tumor fluid flow derived by 2-scale homogenisation techniques in recent works. This biphase fluid flow model accounts for the capillary wall permeability, and the interstitial avascular phase, both being mixed in the limit homogenised problem. When the vessel walls become more permeable, we show that the biphase fluid flow exhibits a boundary layer that makes the computation of the full problem costly and unstable. In the limit, both capillary and interstitial pressures coincide except in the vicinity of the boundary where different boundary conditions are applied. Thanks to a rigorous asymptotic analysis, we prove that the solution to the full problem can be approached at any order of approximation by a monophasic model with appropriate boundary conditions on the tumor boundary and appropriate correcting terms near the boundary are given. Numerical simulations in spherical geometry illustrate the theoretical results.

math.AP↗

arXiv2015

On the growth and dissemination laws in a mathematical model of metastatic growth

Metastasis represents one of the main clinical challenge in cancer treatment since it is associated with the majority of deaths. Recent technological advances allow quantification of the dynamics of the process by means of noninvasive techniques such as longitudinal tracking of bioluminescent cells. The metastatic process was simplified here into two essential components -- dissemination and colonization -- which were mathematically formalized in terms of simple quantitative laws. The resulting mathematical model was confronted to in vivo experimental data of spontaneous metastasis after primary tumor resection. We discuss how much information can be inferred from confrontation of theories to the data with emphasis on identifiability issues. It is shown that two mutually exclusive assumptions for the secondary growth law (namely same or different from the primary tumor growth law) could fit equally well the data. Similarly, the fractal dimension coefficient in the dissemination law could not be uniquely determined from data on total metastatic burden only. Together, these results delimitate the range of information that can be recovered from fitting data of metastatic growth to already simplified mathematical models.

q-bio.TO↗

arXiv2014

Classical Mathematical Models for Description and Prediction of Experimental Tumor Growth

Despite internal complexity, tumor growth kinetics follow relatively simple macroscopic laws that have been quantified by mathematical models. To resolve this further, quantitative and discriminant analyses were performed for the purpose of comparing alternative models for their abilities to describe and predict tumor growth. For this we used two in vivo experimental systems, an ectopic syngeneic tumor (Lewis lung carcinoma) and an orthotopically xenografted human breast carcinoma. The goals were threefold: to 1) determine a statistical model for description of the volume measurement error, 2) establish the descriptive power of each model, using several goodness-of-fit metrics and a study of parametric identifiability, and 3) assess the models ability to forecast future tumor growth. Nine models were compared that included the exponential, power law, Gompertz and (generalized) logistic formalisms. The Gompertz and power law provided the most parsimonious and parametrically identifiable description of the lung data, whereas the breast data were best captured by the Gompertz and exponential-linear models. The latter also exhibited the highest predictive power for the breast tumor growth curves, with excellent prediction scores (greater than 80$\%$) extending out as far as 12 days. In contrast, for the lung data, none of the models were able to achieve substantial prediction rates (greater than 70$\%$) further than the next day data point. In this context, adjunction of a priori information on the parameter distribution led to considerable improvement of predictions. These results not only have important implications for biological theories of tumor growth and the use of mathematical modeling in preclinical anti-cancer drug investigations, but also may assist in defining how mathematical models could serve as potential prognostic tools in the clinical setting.

q-bio.QM↗

arXiv2014

Global Dormancy of Metastases Due to Systemic Inhibition of Angiogenesis

Autopsy studies of adults dying of non-cancer causes have shown that virtually all of us possess occult, cancerous lesions. This suggests that, for most individuals, cancer will become dormant and not progress, while only in some will it become symptomatic disease. Meanwhile, it was recently shown in animal models that a tumor can produce both stimulators and inhibitors of its own blood supply. To explain the autopsy findings in light of the preclinical research data, we propose a mathematical model of cancer development at the organism scale describing a growing population of metastases, which, together with the primary tumor, can exert a progressively greater level of systemic angiogenesis-inhibitory influence that eventually overcomes local angiogenesis stimulation to suppress the growth of all lesions. As a departure from modeling efforts to date, we look not just at signaling from and effects on the primary tumor, but integrate over this increasingly negative global signaling from all sources to track the development of total tumor burden. This in silico study of the dynamics of the tumor/metastasis system identifies ranges of parameter values where mutual angio-inhibitory interactions within a population of tumor lesions could yield global dormancy, i.e., an organism-level homeostatic steady state in total tumor burden. Given that mortality arises most often from metastatic disease rather than growth of the primary per se, this finding may have important therapeutic implications.

q-bio.QM↗