Irene Balelli

Irene Balelli

Tenured research scientist (ISFP - Inria Starting Faculty Position)

Publications

Cardiac Electromechanical Model Sensitivity Analysis using Causal Discovery

Al-Ali S., Rodríguez Padilla J., Sermesant M., Balelli I.

Published in FIMH 2025, 2025

This work firstly uses causal discovery to perform global sensitivity analysis of an electromechanical model, and investigate the interrelationships among input parameters and the main outputs of interest. The obtained causal graph reveals and quantifies the strengths of the relationships among those variables. Those results are further confirmed by independently performing clustering and classification, as well as classical local sensitivity analysis. Our method provides a precise identification of the key parameters to be focused on, and ensures stable results compared to classical global sensitivity analysis methods, despite a limited number of available simulations.

Assessing ion channel blockade and electromechanical biomarkers’ interrelations through a novel Multi-Channel Causal Variational Autoencoder

Al-Ali S., Mora M. T., Sermesant M., Trénor B., Balelli I.

Published in CinC 2024, 2024

Drug-induced Torsade de pointes (TdP) is a critical arrhythmia that can lead to sudden cardiac death. This study aims to employ a novel Multi-Channel Causal Variational Autoencoder (MC2VAE), to identify causal relationships between ionic currents blockades, torsadogenic indices, and electrophysiological biomarkers, considered as three distinct sources of information (channels) for drug-induced TdP risk. We consider the downstream task of drug classification for TdP risk on the latent channels and show a clear improvement of classification performances when combining the three considered channels.

Multi-Channel Causal Variational Autoencoder

Al-Ali S., Balelli I.

Available in HAL, 2024

In this work we propose Multi-Channel Causal VAE (MC2VAE), a causal disentanglement approach for multichannel data, whose objective is to jointly learn modality-specific latent representations from a multichannel dataset, and identify a linear causal structure between the latent variables. We formally derive MC2VAE and the optimization strategy for its parameters. Experiments on synthetically generated data-sets underline the ability of our model to identify ground-truth hidden causal relationships.

A computer model for in-silico trials on pacemaker energy efficiency

Coudière Y., Leguèbe M., Balelli I., Baretta A., Fauré G., Feuerstein D.

Published in CinC 2024, 2024

Pacemakers are commonly required to treat bradycardias. Their capture threshold (minimum energy required to stimulate the heart) is critical to assess and predict pacemaker performance, and may change due to fibrosis associated with the inflammatory process, resulting in loss-of-capture and requiring re-hospitalization. We developed a 3D model that computes threshold curves depending on the pacemaker and cardiac tissue properties. Here, it is used through an in-silico trial to compute the proportion of a population for which the initial device setting no longer captures.

Causality: fundamental principles and tools (Book chapter)

Balelli I., Al-Ali S., Dumas E., Abecassis, J.

Published in MICCAI/Elsevier book series, 2024

This chapter is part of the collection Trustworthy AI in Medical Imaging. Its goal is to provide a gentle introduction to Causal Learning (CL), and motivation for its application to medical image analysis, seeking for more robustness against data and domain drifts, and a reliable tool to answer conterfactuals questions and get improved interpretability. The probabilistic formalism at the basis of CL are itroduced, along with basic definitions and assumptions.

Integration of Multimodal Data (Book chapter)

Lorenzi M., Deprez M., Balelli I., Aguila A.L., Altmann A.

Published in Springer US, 2023

This chapter is part of the collection Machine Learning for Brain Disorders. It focuses on the joint modeling of heterogeneous information, such as imaging, clinical, and biological data. This kind of problem requires to generalize classical uni- and multivariate association models to account for complex data structure and interactions, as well as high data dimensionality.

A causal discovery approach for streamline ion channels selection to improve drug-induced TdP risk assessment

Al-Ali S., Llopis-Lorente J., Mora M. T., Sermesant M., Trénor B., Balelli I.

Published in IEEE Xplore2023, 2023

Torsade de Pointes (TdP) is an extremely serious drug-induced cardiac side effect, which is related to abnormal repolarizations in single cells, and the minimum set of ion channels needed to correctly assess TdP risk is still an open question. In this work, we propose to apply the causal discovery method ICA-Linear Non-Gaussian Acyclic Model (ICA-LiNGAM) to uncover the relations across the 7 ion channels identified by the Comprehensive in vitro Proarrhythmia Assay (CiPA) initiative as potentially related to the induction of TdP.

Fed-MIWAE: Federated Imputation of Incomplete Data via Deep Generative Models

Balelli I., Sportisse A., Cremonesi F., Mattei P. A., Lorenzi M.

Available in HAL, 2023

In this work we propose a novel pipeline for federated preprocessing, based on the deep latent variable model MIWAE for missing data imputation: we show that performing the imputation task taking advantage of the federated network is highly beneficial in term of generalizability and robustness.

Parameter estimation in nonlinear mixed effect models based on ordinary differential equations: an optimal control approach

Clairon, Q., Pasin, C., Balelli, I., et al.

Published in Computational Statistics, 2023

We present a parameter estimation method for nonlinear mixed effect models based on ordinary differential equations (NLME-ODEs). The proposed method aims at regularizing the estimation problem in presence of model misspecifications, practical identifiability issues and unknown initial conditions, using a control theory approach.

A Differentially Private Probabilistic Framework for Modeling the Variability Across Federated Datasets of Heterogeneous Multi-View Observations

Balelli I., Silva S., Lorenzi M.

Published in Journal of Machine Learning for Biomedical Imaging, 2022

This paper is an extension of the work presented at the Information Processing in Medical Imaging 2021 conference: we introduce formal differential privacy guarantees compatibly with the proposed federated Bayesian framework and Expectation Maximization optimization scheme.

A Probabilistic Framework for Modeling the Variability Across Federated Datasets of Heterogeneous Multi-View Observations

Balelli I., Silva S., Lorenzi M.

Published in Lecture Notes in Computer Science, 2021

We propose a new Bayesian framework for federated learning of heterogeneous multi-modal biomedical data. An application to a large database (including medical imaging and clinical scores) on patients with Alzheimer disease shows that the proposed model allows high quality data reconstruction, compared to current auto-encoding methods and federated learning schemes.

Viral rebound kinetics following single and combination immunotherapy for HIV/SIV

Prague, M., Gerold, J. M., Balelli, I., et al.

Available in BioRxiv, 2019

In this paper we conduct a detailed analysis of the kinetics of viral rebound after immunotherapy in the context of HIV infection. We use mathematical models combined with rigorous statistical fitting to quantify the impact of these interventions on viral dynamics, and provide a framework for understanding the relative contributions of different mechanisms for preventing viral rebound and highlight the multifaceted roles of TLR7-agonists for HIV/SIV cure.

Dynamics of the humoral immune response to a prime-boost Ebola vaccine: quantification and sources of variation

Pasin, C., Balelli, I., Van Effelterre, T., et al.

Published in Journal of virology, 2019

In this paper data from three phase 1 trials of a 2-doses vaccine against Ebola virus performed by the EBOVAC1 Consortium are analyzed. A mathematical model based on ODEs is used to characterize the dynamics of the humoral response after the second dose immunization up to 1 year of followup: it allows to estimate the durability of the antibody response and the influence of different factors on the dynamics of the humoral response.

Multi-type Galton-Watson processes with affinity-dependent selection applied to antibody affinity maturation

Balelli, I., Milišić, V., Wainrib, G.

Published in Bulletin of mathematical biology, 2019

Inspired by antibody affinity maturation, the interactions between division, mutation and selection are analysed, assuming that the observed population can be classified according to fitness levels with respect to a target trait. An optimal selection rate has been identified, which allows to maximize the number of high affinity B-cells selected at a given generation.

Mathematical foundations of antibody affinity maturation

Balelli, I.

Available in Hal, 2016

PhD Thesis manuscript. Supervisors: Milišić, V., Wainrib, G., Zaag, H. Antibody affinity maturation is a key process in adaptive immunity : it is a mechanism which allows B-cells to produce high affinity antibodies against a specific antigen. We developed and studied a mathematical framework which allows to pattern the learning process to whom B-lymphocytes are submitted during an immune response. In particular, we model antibody traits as N-length binary strings. Antibody-antigen affinity is naturally characterized using the Hamming distance: therefore the N-dimensional hypercube vertex set defines the affinity landscape of B-cell traits. Our aim is to propose and analyze a mathematical model of the division-mutation-selection process of B-cells during an immune response. Besides the biological motivations, the analysis of this learning process brought us to build a mathematical model which could be relevant to model other evolutionary systems, but also gossip or virus propagation. Our method is based on the complementarity between probabilistic tools and numerical simulations.