Machine learning analysis using 77,044 genomic and transcriptomic profiles to accurately predict tumor type.

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Cancer of Unknown Primary (CUP) accounts for 3-5% of cancers, often diagnosed empirically with poor survival outcomes.
• Standard diagnostic methods struggle to identify the origin of metastatic cancer, especially with low tumor cell percentages.

Purpose of the Study:

• To evaluate the diagnostic accuracy and clinical utility of MI GPSai, a machine learning algorithm using genomic and transcriptomic data, for identifying cancer of unknown primary.
• To assess the impact of MI GPSai predictions on existing diagnoses and its potential to improve patient outcomes.

Main Methods:

• MI GPSai algorithm trained on over 34,000 genomic and 23,000 transcriptomic datasets, validated on nearly 20,000 cases.
• Utilized DNA sequencing and whole transcriptome data coupled with machine learning for tumor type prediction across 21 cancer categories.

Main Results:

• MI GPSai achieved over 94% accuracy in predicting tumor type for 93% of cases, increasing to 97% with the second-highest prediction.
• The assay provided predictions for 71.7% of Cancer of Unknown Primary cases, leading to diagnostic changes in 41.3% upon pathologist review.
• The test also assesses genomic markers essential for targeted therapy selection.

Conclusions:

• MI GPSai offers clinically significant diagnostic information for a substantial proportion of Cancer of Unknown Primary cases.
• Integrating MI GPSai into clinical practice has the potential to enhance diagnostic accuracy and improve patient management.
• The single-test approach maximizes clinical utility by combining origin prediction with essential genomic marker assessment for therapy selection.