print · login   

Multimodal multitarget detection of marine biotoxins using foundation models


Background

Marine biotoxins pose a serious threat to human health, causing severe illnesses such as paralytic, amnesic, diarrheic, and neurotoxic poisoning. One group of such toxins are tetrodotoxins (TTXs). TTX is an extremely potent neurotoxin that, upon human consumption, results in paralysis and respiratory failure. While extensively researched in pufferfish, its occurrence in bivalve mollusks remains poorly understood.

Recent research within our group has identified potential hydrological and meteorological drivers for TTX contamination in bivalve mollusks in the Eastern Scheldt (The Netherlands). While these findings provide valuable insights at the sub-regional scale, predictive performance at a localized (plot-level) scale remains limited. Our preliminary findings suggest that AI – specifically foundation models - on remote sensing (satellite) data can detect marine biotoxins. However, solely remote sensing information lacks environmental context.

Thus, we hypothesize that integrating these data sources, thus creating a multimodal model, can significantly improve localized contaminant prediction. Additionally, we have data on not just TTX but a broad spectrum of marine biotoxins, offering opportunities for multi-toxin modelling.

Research Objectives

The primary objective of the thesis is to develop a multimodal prediction framework of TTX contamination at a plot-level resolution. This will involve the harmonization of multiple data sources, including remote sensing, hydrological, and meteorological datasets. Once such a multimodal framework has been developed, the project can be extended towards a multitask prediction of marine biotoxins. Enabling the joint localized prediction of multiple marine biotoxins. Further research objectives can be tailored to the students’ interests, background, and ambition.

Your role & Requirements:

  • Knowledge: understanding of deep learning
  • Skills: Python programming
  • Interest: depending on research objectives, an interest in remote sensing, multi-task learning, and or multi-modal learning. Furthermore, an interest in biological/contaminant safety.

Practical information & what we offer:

Contact information:

Thijs Schoppema MSc


References:

  1. 2511.20395 Identifying environmental factors associated with tetrodotoxin contamination in bivalve mollusks using eXplainable AI