Delivering A High-Accuracy Predictive Model For Cancer Detection

Delivering A High-Accuracy Predictive Model For Cancer Detection

Table of Contents

Click here to download this case study in PDF

SITUATION

A government-funded cancer center sought to implement an Artificial Intelligence (AI)-powered classifier, utilizing deep neural networks (DNN), for diagnosing cancer using RNA-sequencing data (RNA-seq), but they lacked the specific in-house resources to develop such a complex predictive model, so they turned to Bridge Informatics (BI) for help.

STRATEGY

As a leading bioinformatics data management and analysis company, BI partnered with the Cancer Center to help bring their AI dreams to life. We leveraged our expertise in data mining and deep learning for cancer classification to create a Statement of Work that included the following:

  • Data Acquisition: We accessed cancer-specific bulk transcriptional data from The Cancer Genome Atlas (TCGA) database. We then implemented a filtering process to ensure sufficient patient samples (> 150 patients), with tumor-matched normal tissue, were present for the cancers of interest to train and test the DNN. This resulted in a refined dataset encompassing nine human cancers.
  • Feature Selection: We performed differential gene expression analysis using a trusted algorithm to identify genes exhibiting significant differences between tumor and tumor-matched normal control samples. The resulting gene lists were filtered after carefully choosing a relevant cutoff threshold and acceptable false discovery rate with an appropriately adjusted p-value.
  • Deep Neural Network Development: The resulting dataset, consisting of differentially expressed genes as inputs and cancers as labels, was used for training and testing the DNN architecture. The network architecture consisted of 339 gene inputs, representing the genes passing cutoffs, followed by three hidden layers and proprietary activation functions for complex information processing. Finally, the network culminated in 10 distinct outputs, each corresponding to one of the nine cancer types identified earlier, with an additional output for normal tissue classification.

RESULTS

Within six months, BI equipped the Cancer Center with a powerful deep neural network AI cancer classifier that was trained on the TCGA’s real patient data.

The AI architecture was refined to distinguish cancer from normal tissue with >90% and an F1 score of > 0.8.

The deep neural network, trained on the curated RNA-seq data, is designed to analyze patient samples and predict cancer presence and type with high accuracy, empowering oncologists with valuable insights to support their diagnoses and potentially improve patient outcomes.

Interested in partnering with Bridge Informatics? Contact us to learn more about our team of bioinformaticians with experience at the bench whose core specialty is understanding and analyzing biological data.


Jessica Corrado, Head of Business Development & Commercial Operations, Bridge Informatics

As the Head of Business Development & Commercial Operations, Jessica is responsible for driving strategic growth initiatives and overseeing the company’s commercial activities. She has both a keen understanding of the life sciences industry and a strong track record in building successful partnerships.

Prior to joining Bridge, Jessica held a number of leadership roles across sales, marketing, and communications. Outside of work, Jessica is responsible for the majority of marketing and event planning for Shore Saves, a non-profit animal rescue. She enjoys reading and is often reading at least two books of various genres at a time. If you’re interested in reaching out, please email [email protected] or [email protected]

Share this article with a friend

Create an account to access this functionality.
Discover the advantages