"Synthetic databases accelerate innovation and improve patient privacy"

Ferran Prados is a leading researcher in the field of bioinformatics applied to the brain and member of the NeuroADaS Lab at Universitat Oberta de Catalunya (UOC), a research group that works to advance knowledge of the brain. He has led research published in Nature on the generation of synthetic data to advance research on chronic fatigue. Prados, a member of the UOC's Faculty of Computer Science, Multimedia and Telecommunications, will talk about this project, which won the UOC's Interdisciplinary Research Award, at a Brain Week event that will take place at the university on 12 March.

What is the goal of your research on synthetic data? 

The main objective is to develop methods to generate synthetic health data. In the case of our research, it's data related to quality of life tests for patients with chronic fatigue. These tests require a considerable amount of time to complete, and patients are often unable to answer them in full due to their state of health. The fact that there is little data available for research hinders the development of new treatments, biomarkers and other clinical advances.

Synthetic data can be a valuable solution to help address this challenge. Through a neural network and from already completed tests, it's possible to learn patients' response patterns and generate synthetic responses that are statistically consistent with the real ones. This allows databases to be enriched and the quality of research to be improved without compromising patient privacy.

The results demonstrate that it's possible to generate synthetic datasets that remain useful for research and analysis, while ensuring the protection of individuals' sensitive data.

What benefits does this type of data bring to medical research? 

We don't only generate synthetic databases, we also use them to complete real databases with the aim of improving the sample size and, thus, strengthening the validity of the studies. This approach allows us to optimize the available datasets, especially in cases where there isn't enough information.

Synthetic databases improve privacy by allowing data to be shared without exposing identifiable personal information. They also facilitate wider access to data, which benefits researchers who might otherwise encounter restrictions due to strict data protection regulations. In addition, they contribute to accelerating innovation by providing datasets to train artificial intelligence models, even in situations where real data is scarce, such as in the case of hard-to-access or little-known diseases.

What limitations are there?

The quality of the data is critical, because if the real data used to generate synthetic data is of low quality, the results may not be reliable. So, these models can reproduce biases present in the original data, which can lead to errors or inequalities in the research. Finally, regulations remain a challenge. In some cases, the use of synthetic data in clinical studies isn't looked upon favourably and requires additional validation by the competent authorities before they can be used in clinical practice.

An interesting application of synthetic data is their ability to simulate medical images. 

At NeuroAdAS Lab we work with synthetically generated images to simulate structures such as the optic nerve, teeth and multiple sclerosis lesions. 

This process can be carried out using different techniques. We use Bayesian methods, i.e., methods based on labels to generate synthetic data with a solid physical and morphological basis or generative adversarial networks (GANs) to learn from real medical images and then create new ones while maintaining their essential characteristics. These synthetic images are of great value for the training and optimization of neural networks, and allow them to be more efficient in the segmentation and analysis of medical images.

What artificial intelligence techniques do you use in this research?

There are several, including GANs, where two neural networks compete with each other to generate realistic synthetic data. We also use deep learning, which uses neural networks with multiple layers to model complex patterns in data, and probabilistic models that estimate the distribution of data to generate new data with similar characteristics.

Apart from neurological diseases, is it possible to use these models for other diseases?

Yes, synthetic data generation models can be applied to a wide range of diseases. For example, in the study of cardiovascular, oncological and infectious diseases, among others, to create datasets that help develop new treatments. 

What technological tools do you think will contribute the most to advancing our understanding of the brain?

Advanced artificial intelligence techniques, deep learning applied to neuroimaging, quantum computing and bioinformatics will be crucial to the future of neuroscience. In addition, technologies such as non-invasive neuromodulation, brain-machine interfaces and connectomics (the study of nervous system connections) will allow us to better understand neural networks and develop new strategies for the treatment of diseases.

Do you think bioinformatics can be a game-changer in the diagnosis and treatment of diseases in the future?

Undoubtedly, and it has the potential to become one in the immediate future. Bioinformatics combines biological sciences with computational techniques to analyse and interpret large amounts of biomedical data, and it's become fundamental in current medical research and practice. 

At the UOC we have a team of highly qualified researchers working in precision medicine from different perspectives, with the aim of developing innovative solutions to advance in the diagnosis and personalized treatment of various diseases. Our research focuses on improving the understanding of diseases and offering therapeutic strategies that are adapted to each patient, taking advantage of the potential of bioinformatics and artificial intelligence.

It is also transforming the way new drugs are developed.

This is a key area and the UOC is home to a group that carries out research focused on this field. 

New computational tools make it possible to identify therapeutic targets, simulate the efficacy of new compounds and optimize drug discovery. All this while reducing costs and ensuring that drugs reach patients sooner. 

Likewise, there's genomic analysis, which allows us to identify genetic mutations associated with various diseases so we can detect them early and personalize the treatment the patient receives.

What other projects are you working on that can have an impact on patients?

One project we're working on seeks to improve the segmentation of lesions in magnetic resonance imaging of patients with multiple sclerosis. We've developed an innovative technique that generates synthetic data, simulating the real evolution of lesions when given any image contrast, which then allows us to improve the model. 

Another project focuses on the segmentation of the optic nerve of patients with Multiple Sclerosis, a key structure in the monitoring of the disease and one which has recently been included as a diagnostic factor for the disease. We're developing solutions to make it easier for doctors to interpret its morphology and state.

Another research area we're exploring involves analysing the protective capacity of bilingualism for brain function, using various tests and neuroimaging tools.

Which of your research projects is closest to being integrated into the health system?

Integrating software tools into clinical practice isn't easy due to current regulations. The project on the analysis of lesions in patients with multiple sclerosis will soon be used in international clinical trials, and will help researchers detect and quantify lesions more efficiently, improving diagnosis and personalizing treatments for each patient. 

Let's talk about Brain Week. Why do you think an event of this kind is important?

At the NeuroADaS Lab, we see Brain Week as a key outreach activity to bring neuroscience closer to society, raise awareness about the importance of brain health and inform people of the advances we're making in research. Through this event, we're fostering dialogue between scientists, doctors, and the general public, which is vital to improve people's understanding of neurological disorders and highlight the need to invest in research to improve the quality of life of those affected.