Case studies

Advance Cancer Detection Codvo.ai Leverages DNA-Based Machine Learning for Enhanced Accuracy

index

About the client

Overview

Business Challenge

Solution

Tech Stack

Business Impact

About the client

Our client, a pioneering medical device innovator based in the United States, specializes in revolutionary non-invasive technologies aimed at early cancer detection. They have developed a unique, office-based targeted cell collection device coupled with a sophisticated lab-based DNA biomarker test.

This test is specifically designed to detect precursor conditions for types of cancer, focusing on enhancing the accuracy of cancer risk assessments. By leveraging the potential of DNA methylation levels in key genes, such as VIM and CCNA1, the company is at the forefront of personalized medicine, offering new hope for early and accurate cancer detection.

Overview

The project embarked upon by Codvo.ai with this medical innovator revolved around enhancing the sensitivity of the client's DNA-based data analysis. This enhancement focused particularly on the VIM and CCNA1 genes, known for their critical role in signaling cancer development based on methylation levels. A high percentage of methylation in these genes indicates a higher likelihood of cancer.

Business Challenge

The primary challenge faced by our client was to significantly enhance the sensitivity of their DNA-based data analysis process. This endeavor focused on the accurate detection of precursor conditions for specific cancers, utilizing the methylation levels of the VIM and CCNA1 genes as biomarkers.

Our challenge was to incorporate demographic data such as age, gender, and race to refine benchmark scores, with the goal of achieving over 90% sensitivity and maintaining a high specificity level for reliable cancer risk assessments.

‍

Our Approach and Solution

Identified the challenge as a binary classification problem, focusing on the sensitivity and specificity of detecting cancer precursors through DNA methylation analysis.
Multiple machine learning algorithms were trained, including Logistic Regression, Random Forest, Decision Tree, and SVM, with both linear and non-linear kernels.
Conducted extensive experiments with diverse combinations of demographic data and gene-based data to determine their impact on the algorithms’ effectiveness in decision making.
Employed a rigorous process of algorithm tuning, including hyperparameter adjustments, to optimize model performance.
Achieved significant improvements in the models' ability to accurately assess cancer risks, with more than 90% sensitivity, demonstrating the potential for early detection and personalized patient care.

Tech Stack

Tech used: Python Scikit Learn, Jupyter Notebook, Numpy, Pandas and Matplotlib.

Highlights

Business Impact

Major Outcome

Achieved notable improvements in the sensitivity of DNA-based cancer detection, specifically targeting the VIM and CCNA1 genes.
Logistic Regression and Random Forest algorithms emerged as the best performing algorithm on the validation dataset, reaching sensitivities of 87% and 86% respectively, with both maintaining a specificity of 80%.
Hyperparameter tuning further refined the models, helping us to achieve the goal of minimum 90% sensitivity target.
Codvo's collaboration has set new benchmarks in personalized cancer risk assessment, facilitating early detection and treatment possibilities and illustrating the profound impact of advanced data analysis on healthcare outcomes.

‍