Conformational heterogeneity in photoconvertible fluorescent proteins

Photoconvertible fluorescent proteins are widely used for advanced microscopy techniques. However, their

complex photophysical behaviour limits data quality and quantitative applications. Therefore, understanding

their photophysical properties is essential to optimize experimental conditions and guide the development of improved variants through protein engineering.

I will present our work on the photoconvertible protein mEos4b, performed in collaboration between the NMR and microscopy groups at IBS, uncovering the existence of multiple conformational states that regulate its photophysical behaviours. In the green state, we found two conformations that differ in the protonation state of the E212 and H62 side chains, in a pH-dependent manner, with important consequences for the green-to-red photoconversion efficiency. In the photoconverted red state, we identified a pH-dependent positive photoswitching mechanism that generates a short-lived non-fluorescent state, perturbing single-molecule localization experiments. Together, these findings highlight the complex conformational landscape of fluorescent proteins and its impact on their applications in advanced imaging.

References:

• Light-Induced Conformational Heterogeneity Induces Positive Photoswitching in Photoconvertible Fluorescent Proteins of the EosFP Family. JACS. https://doi.org/10.1021/jacs.4c17311 
• Structural Heterogeneity in a Phototransformable Fluorescent Protein Impacts its Photochemical Properties. Advanced Science. https://doi.org/10.1002/advs.202306272 
• Photophysical studies at cryogenic temperature reveal a novel photoswitching mechanism of rsEGFP2. JACS. https://doi.org/10.1021/jacs.3c01500 

Generative AI and agentic architectures for industrial applications

Artificial intelligence, and particularly generative AI, is rapidly gaining widespread adoption beyond traditional data science domains in industry. However, the use of publicly available AI applications raises important challenges regarding data confidentiality, governance, and integration with company-specific knowledge. At G2M-AI, we develop tailored AI solutions designed to address specific industrial needs while ensuring control over data flows.

I will present several multi-agentic architectures and retrieval-augmented generation (RAG) systems that we have developed and deployed over the past year. These include a company chatbot connected to internal documentation, a RAG-based assistant for drafting technical documents, and a data analysis chatbot capable of generating interactive dashboards. Through these examples, I will illustrate how generative AI and agentic approaches can transform knowledge access, document generation, and data analysis in industrial environments.