Client success story
ML Driven Oncology Diagnostic Development
Business Value Delivered: Valuable Diagnostic With Powerful Clinical Utility
Summary: Machine Learning Used To Develop Cancer Screening Diagnostic
bPrescient worked with a diagnostics and therapeutics company in the oncology space to develop machine learning models to predict tumor tissue of origin from circulating tumor DNA, informing development of a diagnostic product and identification of potential drug targets.
Challenge: Different Cancer Types Difficult To Distinguish Accurately
The company had large internal data sets, and access to public data, of DNA sequences of tumor DNA and circulating tumor DNA from plasma, which are rich sources of diagnostic and drug target information. Markers to identify diagnostic models and drug targets require high levels of specificity and these data sets contain thousands of potential markers. Finding efficient ways to identify these markers is critical for the company to meet its diagnostic product timelines and secure additional funding.
Solution: Multimodal Data and Machine Learning Identified Tissue-Specific Signals
bPrescient’s team, consisting of an AI and Machine Learning (ML) Subject Matter Expert, a ML specialist, and a software and QA engineer, developed and tested statistical and machine learning models on multiple data sets to identify tumor status and tissue of origin. The models initially used sequence variants and haplotypes as input, but were later enhanced to leverage more raw data such as read alignments. The resulting features were reviewed to ensure they were independent of batch effects and multiple iterations of the models led to incremental improvements over time.
Outcome: Identification of Tumor-Specific Diagnostic Model
The predictive models for tumor presence/absence and tissue of origin performed up to the levels required for further development as a diagnostic test, and features identified were also explored for therapeutic purposes. The company was able to secure another round of funding and move forward with their development of a diagnostic product.
Let’s talk
Tell us about your toughest R&D challenge and let’s discuss how bPrescient can solve it.
