Research on AI revolution in genetics

The integration of genomics and transcriptomics has revolutionized the understanding of biological systems by providing comprehensive insights into the relationship between genetic information and gene expression patterns. Traditionally, genomics and transcriptomics have been studied independently. Still, recent advancements in multi-view approaches have enabled the simultaneous analysis of genomic and transcriptomic data, leading to a more holistic understanding of cellular processes. Our main interests concern multi-view methods leveraging the complementary information encoded in genomic and transcriptomic datasets to uncover novel biological insights. Computational and statistical techniques employed in multi-view integration involve co-expression network analysis, clustering algorithms, and dimensionality reduction methods. The benefits of multi-view integration include the identification of regulatory mechanisms, the discovery of novel biomarkers, and the elucidation of complex gene regulatory networks. Integrating genomics and transcriptomics through multi-view approaches represents a powerful paradigm shift in biological research, enabling comprehensive analyses and opening new avenues for discoveries at the intersection of genetic variation and gene expression.

Gene fusions, resulting from genomic rearrangements, have emerged as important drivers of oncogenesis and potential therapeutic targets in various types of cancer. Detecting gene fusions accurately is crucial for understanding tumor biology and developing targeted therapies. In recent years, advancements in RNA and DNA-based approaches have significantly improved gene fusion detection capabilities. 

Our aim is to unravel genomic rearrangements through RNA and DNA-based gene fusion detection tools, which are promising for advancing precision medicine in oncology and improving patient outcomes.

MicroRNAs (miRNAs) are small non-coding RNA molecules that play critical roles in post-transcriptional gene regulation. Identifying the targets of miRNAs is crucial for understanding their functional implications in various biological processes and diseases. Computational prediction approaches have emerged as valuable tools for predicting human miRNA targets, as experimental methods can be laborious and time-consuming. We are interested in computational prediction approaches for human miRNA target identification, providing algorithms and databases that leverage sequence complementarity and evolutionary conservation principles to predict miRNA-target interactions. As computational prediction approaches continue to improve, they provide valuable insights into miRNA target interactions, enhancing our understanding of gene regulation and the intricate mechanisms underlying human biology and disease.

Medulloblastoma is a malignant brain tumor primarily affecting children, necessitating continuous advancements in research to improve diagnosis, prognosis, and treatment strategies. In recent years, integrating genomics and proteomics has emerged as a powerful approach for gaining comprehensive insights into the molecular landscape of medulloblastoma. Medulloblastoma research has made significant strides toward personalized medicine approaches and targeted therapies by leveraging genomics and proteomics insights from sample analysis. However, challenges remain, including data integration, validation, and translation into clinical practice.

This project is developed with the “Signaling in the Development and Brain Tumor” Lab at the Institut Curie Center, Orsay, Paris. Nonetheless, the advancements achieved through genomics and proteomics analyses of medulloblastoma samples have propelled the field forward, bringing us closer to improved patient outcomes and, ultimately, a cure for this devastating disease.

Understanding gene expression is crucial for deciphering cellular processes and disease mechanisms. Traditional methods primarily focus on the analysis of DNA sequence data to predict gene expression levels. However, recent research has highlighted the importance of post-translational modifications (PTMs) in regulating gene expression. PTMs, such as methylation, play key roles in modulating protein activity and stability, ultimately impacting gene expression. 

By incorporating PTM data, researchers can gain a deeper understanding of the regulatory mechanisms governing gene expression dynamics. Our aim is to encompass computational and experimental approaches that leverage DNA sequence data and PTM profiles to predict and interpret gene expression patterns. Decoding gene expression through the synergy of DNA sequence analysis and PTM exploitation provides a powerful framework for unraveling the complexities of gene regulation and its impact on cellular function and disease processes.

The increasing amounts of biological and patient clinical data and the rapid pace of associated research results offer new opportunities for improved treatment decisions but also increase the complexity of the decision-making process. The Oncodash software addresses this problem by assisting clinicians with patient review and the prioritization of treatment options. OncoDash provides interactive interfaces on not only patient data but also a rich environment to interact with models. In particular, it includes AI-driven models, which show surprising levels of performance in health and across other applications.

Recent advancements in histopathology and genetics have enabled the development of personalized medicine approaches that can improve patient outcomes by tailoring treatments based on individualized genetic and clinical characteristics. In the field of oncology, chemotherapy is often used as a treatment option, but the efficacy of chemotherapy varies between patients due to differences in genetic makeup and tumor characteristics. Therefore, identifying predictive biomarkers of chemotherapy response is crucial for effective treatment planning. In this context, we integrate histopathology images with genetic features aimed at predicting chemotherapy response in ovarian cancer patients. Ultimately, the convergence of histopathology and genetics has the potential to transform cancer treatment by enabling the development of more personalized and effective treatment plans based on an individual patient's unique molecular profile.

The project aims to develop deep-learning techniques for analyzing biological images and videos. The study of the interaction of bacteria with the host is an integral part of vaccine design. It is usually performed using cell models and ad-hoc microscopy, often with non-scalable approaches that drastically reduce the throughput of the methodology and its application for strong antimicrobial resistance. To evaluate antibodies' effectiveness against bacterial strains, we are developing methods combining image and video analysis and machine learning/deep learning techniques. The project’s goals fulfill the national need for protocols and intelligent tools for discovering therapy resistance mechanisms and new molecules against infectious agents for developing monoclonal antibodies and vaccines.