PhD position Untangling multi-property NMR signals in drug screening with data-driven neural networks

Einleitung

The Scientific Computing Center (SCC) is the Information Technology Center of KIT.

The junior research group “Robust and Efficient AI” at SCC conducts research on scalable AI methods for applications in the natural sciences. The team focuses particularly on the question of how machine learning can be made more robust and efficient to enable the use of such methods in complex and safety-critical application areas.

Since many of these applications rely on extremely large datasets, high-performance computing (HPC) plays a central role in the group’s research.

Curious about an exciting and versatile role in an agile team? Discover more about SCC as your professional place to be: https://www.scc.kit.edu/en/aboutus/working-at-scc.php

Aufgaben

Within the Collaborative Research Center (SFB) HyPERiON at KIT, an innovative PhD project is offered that focuses on resolving signal overlap in parallel NMR spectroscopy using artificial intelligence (AI). NMR spectroscopy is a key tool in drug discovery. However, in a parallel setup, signal couplings and overlaps occur that make it difficult to extract critical molecular information. The aim of the project is to develop AI models capable of generating individual, decoupled spectra from coupled NMR spectra.

Within the scope of the project, your responsibilities will include:

• Developing a transformer-based neural network for the processing of NMR spectra
• Creating datasets from existing experiments within the CRC and from your own experiments, which are to be carried out during a research stay at KIT’s Institute of Microstructure Technology (IMT)
• Applying self-supervised pretraining based on masked sequence modeling and task-specific fine-tuning to the trained neural network
• Analyzing the extent to which the developed model can learn the underlying physical principles of nuclear magnetic resonance

You will further be part of HyPERiON, participating in CRC activities and engage with the other PhD students and projects.

Key Focus Areas

• Scalable deep learning methods for nuclear magnetic resonance
• Self-supervised pre-training techniques and transfer learning approaches in Transformer-based architectures
• GPU-based computing and high-performance computing (HPC)
• Application of AI methods in a scientific context

Profil

Job requirements:

• M.Sc. in computer science, physics, mathematics or equivalent discipline
• Very good programming and software development skills, preferably in Python
• Prior experience with deep learning model development and training, or nuclear magnetic resonance methods