Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Concept-guided graph representation learning of tumor microenvironment
In IBM Research AI4ScR team, our focus is on understanding the spatiotemporal heterogeneity of biological organization. To achieve this, we combine AI, especially Deep learning, with spatial single-cell measurements for modeling the complexity of tumor microenvironment (TME).
Graph representation learning has recently become popular in single-cell biology to understand the spatial cell distribution in TME [1,2,3,4,5]. Encoding the spatial arrangement of cells in form of a graph allows to capture the inter-cellular interactions and map the topological distribution to downstream biological metadata. Though the downstream performance for using cell-graphs is encouraging, the explainability of the findings in terms of the functional capabilities of the cells is missing.
To bridge this gap, the project aims at exploring a novel concept-paradigm, that builds on prior biological knowledge to construct explainable representation of TME. The paradigm is expected to identify various entities of interest in tissue, i.e., cells, and engineer entity-guided abstract concepts to encode the TME. Subsequently, the multitudes of information in terms of the designed concepts are to be integrated via attention-based Graph Neural Networks (GNNs) for tumor characterization. Eventually, the objective is to decipher the learned networks via post-hoc interpretability techniques [6] to comprehend the concept-space.
Keywords: Graph Neural Network
Interpretability
Tumor microenvironment analysis
Multiplex Imaging Mass Cytometry
Single Cell Computational Biology
• Literature review of graphs in Computational Biology. • Graph-based concept-space engineering. • Develop an attention-based GNN framework to map the concept-space to downstream task. • Post-hoc techniques to interpret the GNN framework. • Perform ablation study and evaluation on multiple Imaging Mass Cytometry datasets.
• Working knowledge of Python. • Strong theoretical understanding of computer vision, machine learning, and deep learning. • Practical experience with PyTorch. • Knowledge of Graph Neural Networks (GNNs) is preferred.
Not specified
IBM is committed to diversity at the workplace. With us you will find an open, multicultural environment. Excellent flexible working arrangements enable both women and men to strike the desired balance between their professional development and their personal lives.
Starting: September-October, 2022
Starting: September-October, 2022
Please contact Pushpak Pati (pus@zurich.ibm.com), including your CV.
NOTE: The position is only for ETH master thesis students.
Please contact Pushpak Pati (pus@zurich.ibm.com), including your CV. NOTE: The position is only for ETH master thesis students.