Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session C

Poster A3 - Use of Data Integration and Machine Learning to Identify Cancer Biomarkers, Aiding in Diagnosis and Treatment

Friday, October 25, 2024
10:00 AM - 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Colorectal cancer (CRC) is the third most common cancer, accounting for about 10% of all cancer cases. Twenty-three percent of CRC cases are found after distant spreading. While the treatment prognosis looks generally good for those diagnosed before spreading or spreading only to regional areas, those found with spreading to distant organs have only a 15.6% chance of survival after five years. For those with a poor prognosis, time is of the essence. However, there are many treatment combinations, and with chemotherapy alone, there are several drug combinations. Each treatment lasts an average of 3-6 months. It is unlikely that the drug combination that works best for a particular patient is found on the first try.

Machine learning (ML) and AI have recently made waves in the healthcare industry. One promising application is AI/ML-driven genomic data analysis. By visually and graphically representing genomic data, patterns and biomarkers can be identified. CRC is an excellent candidate for such technology because the microbes are largely oral in origin, and not from genetics. Using ML to predict cancer stages from the tumor microbiome also allows for more accurate cancer treatments. For example, perhaps a brain tumor has been found; upon analyzing the tumor microbiome with ML, we discover that the brain tumor has a microbiome that matches that of CRC. Now, it is clear that either the patient has a history of CRC or that the original tumor was missed and that the cancer has metastasized to the brain.

Materials and

Methods:  Data for this research was obtained from The Cancer Genome Atlas Program (TCGA) database and then analyzed using Python. First, the data was filtered. The entire data set contained about 10,000 patient data entries from multiple cancer types. The data set was filtered down to the following TCGA project names: “COAD,” “SARC,” and “READ.” These three projects contained cancers of the rectum, colon, and rectosigmoid junction. After filtering, the data set contained 593 entries. Cancer stage and microbes with their abundances were utilized from the data set. The data set contained a total of 1033 microbes and their respective abundances. Linear Discriminant analysis was used for dimensionality reduction before training the data. K-fold cross-validation was also utilized to ensure no overfitting by using five k-folds. A neural network called a Multi-layer Perceptron algorithm was used to train the machine learning model using a hidden layer size of 100.

Results, Conclusions, and Discussions: Using the data analysis components of Python, the top 20 microbes per cancer type were identified. The 5-fold cross-validation accuracy scores consistently fall within 97-99%. The latest run had accuracy scores of: 0.99159664, 0.99159664, 0.98319328, 0.97457627, and 0.94915254. The accuracy had a standard deviation of 1.57%, and the weighted and macro averages were 99%. The precision ranges from 97-98%, the recall ranges from 99-100%, and the F1 score ranges from 98-99%.

Thorough research of CRC and other colon cancers revealed that the microbes associated with the TME are the following: Bacteroides, Streptococcus, Fusobacterium, Clostridium, Prevotella, Faecalibacterium, Blautia, Alistipes, and Enterococcus. These known microbes comprise 8 of 20 microbe genera in CRC Types 1, 2, and 4. For CRC Type 3, they make up 9 of 20. This shows that 40-45% of these microbes are easily identifiable in current research. However, the rest are not and may possibly be a new discovery. Identifying these important microbes will pave the way for this same data analysis for other cancers. It may lead to even more discoveries of microbes associated with specific cancers and their respective types. The machine learning model effectively predicts cancer stages based on the microbes present. Implementing this technology with rectal exams can provide fast and reliable results for any biopsies taken. With the addition of effective treatment per microbe, the future of this machine learning model will also conclude the best treatment for individual patients based on their TME, which will effectively improve prognosis.

Machine learning and coding provide an excellent framework for identifying cancer biomarkers. The future of this machine learning model will use cancer treatments and how they affect the individual microbes to predict the best treatment per patient. Use of the discoveries concluded from this research will contribute to better prognosis and treatment of cancer across many types of cancer. With the addition of effective treatment per microbe, the future of this machine learning model will also conclude the best treatment for individual patients based on their TME, which will effectively improve prognosis.

Acknowledgements (Optional):