Sigma Xi 2021 Andrew Gao Presentation
- 1. Machine Learning and Bioinformatics Approach Yields Noninvasive miRNA Biomarkers for Early Lung Cancer Detection Andrew Gao 1
- 2. Abstract Non-small cell lung cancer (NSCLC), the most common type of lung cancer, affects millions of people. In 2020, lung cancer caused 1.8 million deaths, partly because it is difficult to diagnose lung cancer at early stages. Detecting cancer earlier results in better survival. One potential method for early diagnosis, liquid biopsy, relies on biomarkers present in body fluids. MicroRNAs (miRNAs) in blood could serve as biomarkers for noninvasive diagnostic/prognostic tests. MiRNAs are RNA molecules that regulate the expression of specific genes. Through differential expression analysis of four public datasets, the present study identified 13 miRNAs that are consistently underexpressed in the tissue, blood, and serum of NSCLC patients. Kaplan-Meier survival analysis found that six miRNAs had statistically significant prognostic power (miR-140, miR-199a, miR-29c, miR-320e, miR-103a, miR-526b). Functional enrichment analysis of the genes targeted by the miRNAs demonstrated that they are involved in several hallmarks of cancer, such as the epithelial to mesenchymal transition. A machine learning model was constructed using microRNA expression data. Recursive feature elimination was performed to select miRNAs with the greatest diagnostic value. Five classifiers were tested on the selected miRNAs, with Random Forest and Logistic Regression performing the best. A novel three-microRNA panel with 91.5% accuracy for NSCLC detection was identified (miR-320e, miR-103a, miR-526b). These miRNAs also have significant prognostic power for lung adenocarcinoma. The machine learning and analysis workflow was adapted into an open-source online tool for automatic biomarker selection, available at biomarkergenie.com. Future steps include experimental validation and clinical trials. 2
- 3. Background • 1.8 million deaths from lung cancer in 2018 – 5 year survival rate is only 21% • My grandfather passed away at 67 due to non-small cell lung cancer • Hard to accurately diagnose lung cancer – CT Scan – Tissue Biopsy (follow up) • Cells release molecules like DNA and RNA into the bloodstream • Drawing blood is noninvasive • Different levels of biomarkers could indicate cancer – High levels of miR-1228-3p (Xue 2020) 3 Decreased gene expression (degraded miRNA can’t be translated into protein)
- 4. Statement of Purpose 4 Purpose: Identify non-invasive microRNA biomarkers for non-small cell lung cancer (NSCLC) diagnostic and prognostic tests. Criteria: 1. Must be found in blood 2. Consistently differentially expressed across studies ○ Account for high variance in results between studies 3. Statistically significant change 4. Ideally has biological relevance Outline: 1. Find miRNAs that are reliably differently expressed in lung cancer 2. Find what pathways these miRNAs are involved in 3. Check if these miRNAs can predict survival 4. Identify best combination of miRNAs for diagnosing lung cancer
- 5. Materials: ● Computer ● R Studio ○ R ○ Limma ● Google Colab ● Atom code editor ● Python ○ Matplotlib ○ Seaborn ○ Scikit Learn ○ Streamlit ● Jvenn ● Heroku ● Github ● Gene Expression Omnibus (GEO) ○ GSE137140 ○ GSE93300 ○ GSE94536 ○ GSE53882 ● The Cancer Genome Atlas (TCGA) ● Search Tool for the Retrieval of Interacting Proteins (STRING) ● Kaplan Meier Plotter (kmplot.com) ● Graphpad Prism 9 ● Cytoscape ○ MCODE ● Gene Ontology ○ Panther ● miRWalk ● miRDB ● miRTarBase ● GeneCards Data: miRNA expression profiling data Control: Non-cancer people Experimental: NSCLC patients Procedure:
- 6. 6 Selected datasets and characteristics Differential expression data ● Four datasets were selected. ● In total, 1978 NSCLC and 1932 control samples. The limma package in R was used to calculate logFC and p values (t-test). cutoff: p<0.05 logFC: ratio of expression in disease vs. controls negative = underexpressed positive = overexpressed microRNA name ratio of expression other name
- 7. 7 Venn diagram of overlapping differentially expressed microRNAs between datasets Heatmap of logFC values of each of the 13 microRNAs across all datasets 13 microRNAs are differentially expressed in all four datasets (p<0.05). All are underexpressed (negative logFC).
- 8. 8 349 target genes are involved in many hallmarks of cancer: ● epithelial to mesenchymal transition ● SMAD protein phosphorylation ● heterochromatin ● miRNA silencing (impaired) ● transforming growth factor beta Protein protein interaction (PPI) network of target genes Highly interconnected clusters of interacting genes (using Cytoscape)
- 9. 9 Kaplan Meier Survival Analysis: miR-140, miR-199a, miR-29c, miR-320e, miR-103a, and miR-526b can reliably predict patient survival outlook (p<0.05) high expression = better survival (this makes sense) Squamous Carcinoma Adenocarcinoma
- 10. Can these miRNAs distinguish lung cancer and healthy controls? 10 Input raw data Stage characteristics of input data (GSE137140) Most samples are Stage 1 (72%) columns: microRNAs, target = label ( 1 for cancer, 0 for control) Initial 3 component PCA using 13 miRNAs shows distinct separation between lung cancer and healthy 63% explained variance
- 11. Which classifier works the best? 11 Extra Forest, Random Forest, and Recursive Feature Elimination (Logistic Regression) rank 13 miRNAs by importance ● Generally agreed Random Forest performs best 20% of data for training 80% for testing Which miRNAs are most important? Top 3: miR-320e, miR-103a, miR-526b Chart of Feature Importances based on Extra Forest Ranker
- 12. 12 Confusion matrix of results for top 3 microRNAs 10 fold cross validation on top 3 miRNAs: Accuracy = 0.915 (stdev = 0.13) 10 fold cross validation on top 4 miRNAs: Accuracy = 0.916 (stdev = 0.13) (virtually identical) Accuracy: miR-320e: 82.5% top 3: 91.6% top 4: 92.6% Which combination of miRNAs works the best?
- 13. biomarkergenie.com 13 “one click” omics data analysis tool simply input a csv file works on: metabolomics, molecular descriptors, etc.
- 14. Conclusion Combined machine learning/bioinformatics approach identifies: • 2 miRNA panel with 90% accuracy – miR-320e + miR-103a • 3 miRNA panel with 91.5% accuracy – miR-320e + miR-103a + miR-526b – prognostic biomarkers for squamous carcinoma • 3 miRNA panel with 86% accuracy – miR-140 + miR-199a + and miR-29c – prognostic biomarkers for adenocarcinoma • Potential “2 in 1” tests Web tool: • Speeds up exploratory analysis • No setup needed • Increases accessibility of machine learning • Widely applicable to any disease or omics data 14
- 15. Conclusion Advantages • biomarkers were tested on majority Stage 1 patient data • consistent across large sample size and across 4 studies – methodology of this study accounts for variance between studies • greater sampling flexibility (microRNAs are present in serum AND plasma) Limitations: • small sample size for some data • groups LUSC and LUAD together • flaws with training data – batch effect/confounding Next steps • Experimental validation • Differences between LUAD and LUSC • Clinical trials • Add functionality to web tool – Regression Topics: 1. microRNAs in cancer 2. differential expression analysis 3. target gene prediction 4. functional enrichment analysis 5. machine learning classification 6. biomarkergenie.com
- 16. References 1. Press Release N° 263. (2018). In World Health Organization. International Agency for Report on Cancer. 2. American Cancer Society. Facts & Figures 2019. American Cancer Society. Atlanta, Ga. 2019. Howlader N, Noone AM, Krapcho M, Miller D, Bishop K, Kosary CL, Yu M, Ruhl J, Tatalovich Z, Mariotto A, Lewis DR, Chen HS, Feuer EJ, Cronin KA (eds). SEER Cancer Statistics Review, 1975-2014, National Cancer Institute. Bethesda, MD, https://seer.cancer.gov/csr/1975_2014/, based on November 2016 SEER data submission, posted to the SEER web site, April 2017. 3. Heneghan, H. M., Miller, N., & Kerin, M. J. (2010). MiRNAs as biomarkers and therapeutic targets in cancer. Current Opinion in Pharmacology, 10(5), 543–550. 4. Farazi, T. A., Spitzer, J. I., Morozov, P., & Tuschl, T. (2010). miRNAs in human cancer. The Journal of Pathology, 223(2), 102–115. 5. Ma, J., Lin, Y., Zhan, M., Mann, D. L., Stass, S. A., & Jiang, F. (2015). Differential miRNA expressions in peripheral blood mononuclear cells for diagnosis of lung cancer. Laboratory Investigation, 95(10), 1197–1206. 6. Hennessey, P. T., Sanford, T., Choudhary, A., Mydlarz, W. W., Brown, D., Adai, A. T., Ochs, M. F., Ahrendt, S. A., Mambo, E., & Califano, J. A. (2012). Serum microRNA Biomarkers for Detection of Non-Small Cell Lung Cancer. PLoS ONE, 7(2), e32307. 7. Shen, J., Todd, N. W., Zhang, H., Yu, L., Lingxiao, X., Mei, Y., Guarnera, M., Liao, J., Chou, A., Lu, C. L., Jiang, Z., Fang, H., Katz, R. L., & Jiang, F. (2010). Plasma microRNAs as potential biomarkers for non-small-cell lung cancer. Laboratory Investigation, 91(4), 579–587. 8. Heneghan, H. M., Miller, N., Lowery, A. J., Sweeney, K. J., Newell, J., & Kerin, M. J. (2010). Circulating microRNAs as Novel Minimally Invasive Biomarkers for Breast Cancer. Annals of Surgery, 251(3), 499–505. 9. Xue, W.-X., Zhang, M.-Y., Rui Li, Liu, X., Yin, Y.-H., & Qu, Y.-Q. (2020, July 7). Serum miR-1228-3p and miR-181a-5p as Noninvasive Biomarkers for Non-Small Cell Lung Cancer Diagnosis and Prognosis. BioMed Research International. 10. Ying, Lisha, et al. "Development of a serum miRNA panel for detection of early stage non-small cell lung cancer." Proceedings of the National Academy of Sciences 117.40 (2020): 25036-25042. 11. Chugh, P., & Dittmer, D. P. (2012). Potential pitfalls in microRNA profiling. Wiley Interdisciplinary Reviews: RNA, 3(5), 601–616. https://doi.org/10.1002/wrna.1120 12. Tang, Gusheng, et al. "Different normalization strategies might cause inconsistent variation in circulating microRNAs in patients with hepatocellular carcinoma." Medical science monitor: international medical journal of experimental and clinical research 21 (2015): 617. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4345856/ 13. Kenny, Louise C., et al. "Novel biomarkers for pre-eclampsia detected using metabolomics and machine learning." Metabolomics 1.3 (2005): 227-234. https://link.springer.com/article/10.1007/s11306-005-0003-1 14. Huang, Yao, et al. "Serum microRNA panel excavated by machine learning as a potential biomarker for the detection of gastric cancer." Oncology reports 39.3 (2018): 1338-1346. https://www.spandidos-publications.com/10.3892/or.2017.6163 16 Full list in notebook.
- 17. Acknowledgement Thank you to my parents, my teachers, the Van Allen lab, and my mentor Wendy Slijk for their support! 17