Complete scripts used to produce the results of Imperial MRes Biomedical Research - Data Science 2024-2025 Project 1.
In the thesis, we present a novel multi-omics data integration framework that consists of two steps: first, converting molecular statistics to pathway statistics; and second, constructing and decomposing a tensor using the pathway statistics generated from the first step to extract biological insights. We aim to demonstrate that this integrated framework benefits from the biological knowledge base of pathway analysis and the intuition behind tensor decomposition algorithms. We leverage PARAFAC I and PARAFAC II, two tensor decomposition algorithms, to develop and compare tensors. We build three models based on the two decomposition methods to evaluate model performance. Specifically, we aim to investigate:
- To which level the decomposed results can be interpreted from a biological standpoint.
- Whether the tensor structure effectively accounts for cross-omics interdependencies. And
- Whether the shape flexibility embedded in PARAFAC II leads to more enriched biological signals compared to PARAFAC I.