Mahawan, Tanakamol 
ORCID: 0000-0002-9406-8350
(2024)
Biomarker Identification of Pancreatic Ductal
Adenocarcinoma Metastasis by Systems Biology Approach
Doctor of Philosophy thesis, University of Liverpool.
|
Text
201531057_Sep2024.pdf - Author Accepted Manuscript Download (7MB) | Preview |
Abstract
Systems biology synergises biology with computational sciences, offering a holistic perspective and cutting-edge tools like machine learning (ML) to reveal disease biomarkers – molecular fingerprints for detection, prognosis, and therapeutics. ML can help to identify key molecular signals from omics data that can be used for biomarker candidate discovery. However, effective ML demands practices avoiding data leakage and enhancing reproducibility. In this study we aimed to build a robust pipeline for the discovery of biomarker candidates using machine learning and applied it to PDAC. PDAC is an aggressive pancreatic cancer with a poor prognosis. Over 80% of patients are diagnosed at late metastatic stages with no standard markers known for rapid and cost-effective metastasis diagnosis While many studies focus on early PDAC detection, which is challenging, biomarkers for metastasis remain elusive, underscoring an urgent need to discover metastatic biomarkers and improve the treatment. In this work, we have reused and integrated RNA sequencing (RNA-Seq) data from five public repositories, yielding 489 patient samples. A robust pre-processing pipeline implemented batch effect correction using ARSyN to enable unbiased multi-study data integration. The analytical workflow utilised an N-fold cross-validation approach for variable selection implementing three methods: LASSO, Boruta and VarselRF for robust variable selection from training data only to prevent leakage, yielding 15 candidate genes. A resampling strategy by ADASYN addressed class imbalance common in clinical data.The random forest (RF) model, trained solely on the 15 genes, demonstrated exceptional metastasis classification model performance during internal, cross-dataset, and pseudo-external validation. This suggests that the analysis workflow effectively prevented data leakage and overfitting. Biological contextualisation through QIAGEN IPA and GeneMania revealed complex interplay among the 15 genes in immune response, ECM remodelling, metabolic reprogramming, and angiogenesis – hallmark PDAC metastasis processes. Some genes showed therapeutic target potential, as evidenced by many cancers. Limitations included data integration challenges (with that one step through batch correction for spatial integration of data), low tumour cellularity potentially masking PDAC signals from samples and lack of comprehensive protein data for validation. Future work could explore single-cell RNA sequencing (scRNA-Seq) for cellular insights, studying the TME, therapeutic targeting studies and mapping biomarkers to PDAC subtypes for precision medicine. and wet lab validation for confirming the biomarker candidates.Although focused on PDAC, the general workflow can be applied to other biological contexts exploiting omics data. Each promising finding makes us closer towards a future where cancers will not be death sentences and will become curable diseases.
| Item Type: | Thesis (Doctor of Philosophy) |
|---|---|
| Divisions: | Faculty of Health & Life Sciences > Inst. Systems, Molec & Integrative Biology > Inst. Systems, Molec & Integrative Biology |
| Depositing User: | Symplectic Admin |
| Date Deposited: | 01 Oct 2024 11:24 |
| Last Modified: | 08 Feb 2025 03:06 |
| DOI: | 10.17638/03184261 |
| Supervisors: |
|
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3184261 |
| Disclaimer: | The University of Liverpool is not responsible for content contained on other websites from links within repository metadata. Please contact us if you notice anything that appears incorrect or inappropriate. |
Altmetric
Altmetric