Our scientific publications

Octopize's Avatar method generates synthetic data that maintains all the statistical richness of real data, while guaranteeing the confidentiality of individuals. Designed according to the criteria of the European Data Protection Board, it offers a patient-centered approach and robust anonymization, without compromising between usefulness and protection.

Chronos is an anonymization method dedicated to temporal data, tested on ECG signals. It generates synthetic data that is faithful to the original, with only a 3% difference in accuracy on the trained models. An approach that protects individuals while maintaining the scientific quality of data.

This study introduces ELSTM, a neural network that analyzes patients' complete histories to more accurately predict their risk of mortality at one year after admission. To promote research while maintaining confidentiality, the authors have also published a synthetic data set reproducing the real characteristics of patients.

This article presents the genetic analysis of the OFSEP-HD cohort, comprising more than 2,600 patients with multiple sclerosis followed over five years. The study highlights the genetic diversity of patients and the limits of original self-declarations, while proposing the creation of a synthetic and anonymous genetic data set to promote sharing and open research.

This article discusses the use of the Avatar method to generate synthetic randomized clinical trials using real data from patients with multiple sclerosis. The results show that this approach makes it possible to accurately reproduce clinical analyses while guaranteeing robust confidentiality, even in the face of the most complex inference attacks. Two sets of anonymous and open placebo data have thus been published to demonstrate the potential of the secure sharing of health data.

This article presents a large-scale scientific validation of the Octopize Avatar method, applied to the CONSTANCES cohort (more than 160,000 individuals). The results show that the synthetic data faithfully reproduce the statistical associations of real data, with minimal differences and a high degree of agreement between the results. This study confirms that avatar data offers a reliable and GDPR-compliant alternative for sharing and exploiting massive data safely.

This article presents the French ICAN project, a national program aimed at better understanding the mechanisms of formation and rupture of intracranial aneurysms. By combining high-throughput genetic analyses, imaging and clinical data, the project seeks to identify new genes and biomarkers that are predictive of aneurysm risk. This work paves the way for the development of new diagnostic and therapeutic tools for better prevention of aneurysms.

This article presents the collaboration between Octopize and the Health Data Hub around an educational notebook dedicated to the generation and evaluation of synthetic health data. This work compares several approaches to measure both the quality and the level of anonymity of the data produced. The Octopize method is distinguished by its ability to guarantee confidentiality while maintaining scientific reproducibility, regardless of the use case.

This article compares three synthetic data generation methods — CT-GAN, TVAE, and Avatar — applied to a set of pharmacogenetic data. The results show that Avatar (k = 10) and CT-GAN offer the best balance between data usefulness and privacy, with estimates close to real data. This study highlights the potential of synthetic data for pharmacological research, while identifying ways to optimize small data sets.