Abstract 7133: Detection of early stage colorectal cancer using cell-free oncRNA biomarkers and AI

Abstract Background: Colorectal cancer (CRC) is the second leading cause of cancer-related mortality worldwide. Early detection offers the best opportunity for effective treatment. Although CRC screening methods exist, there remains a need for a high performing blood test to both improve adherence and detect early stage disease. We have previously identified and developed a novel category of cancer-associated, small orphan non-coding RNAs (oncRNAs), and combined them with generative AI modeling to develop a liquid biopsy platform. This platform has demonstrated high sensitivity and specificity for detection of early stage disease across several types of cancer. In this study, we developed an oncRNA- and AI-based assay and analyzed it in a separate, independent test set to evaluate its generalizability and its ability to detect CRC across different cancer stages. Methods: We utilized The Cancer Genome Atlas (TCGA) small RNA profiles to discover a library of pan-cancer oncRNAs that were significantly enriched among tumors compared to adjacent normal tissues spanning multiple tissue sites. The diagnostic performance of these oncRNAs was evaluated in plasma using a training and independent test cohort. Our training cohort included 613 samples (388 CRC and 225 asymptomatic controls) from six different sources (mean age: 63.7 years, 35.9% female). 37.6% of the cases were classified as stage I/II and 60.8% as stage III/IV. Further, we acquired 192 samples (113 CRC and 79 asymptomatic controls) from a separate cohort for testing (average age: 62.6 years ; 51.6% female). In the test set, 52.2% of cases were classified as stage I/II and 47.8% as stage III/IV. We processed 1 ml of plasma for each sample through our universal, automated cell-free RNA workflow and sequenced at an average depth of 58 million 100-bp single-end reads. A generative AI model was trained using 5-fold cross-validation (CV) to predict CRC, and then applied on the separate, independent test set. Results: Our oncRNA-based generative AI model achieved an overall AUC of 0.93 (95% CI: 0.91-0.95) for prediction of CRC versus cancer-free controls in CV, and an overall AUC of 0.95 (0.93-0.98) in the independent test set. At 90% specificity, overall model sensitivity was 81.7% (77.5%-85.4%) by CV in the training data, and 88.5% (81.1%-93.7%) in the test set. For stage I CRC, the model has a sensitivity of 72.1% (59.9%-82.3%, N=68) by CV in the training data, and 80.0% (51.9%-95.7%, N=15) in the test set, both at 90% specificity. Conclusions: We demonstrate that blood-based cell- free RNA can be used for accurate and generalizable early stage detection of CRC. This approach, leveraging a distinct RNA biomarker, has the potential to overcome performance plateaus seen with conventional DNA-based methods in early-stage CRC detection. It offers a promising alternative for blood-based CRC screening in patients and could complement existing DNA- and methylation-based platforms. Citation Format: Amir Momen-Roknabadi, Mehran Karimzadeh, Nae-Chyun Chen, Taylor B. Cavazos, Jieyang Wang, Jeremy Ku, Alex Degtiar, Akshaya Krishnan, Martha Hernandez, Alice Huang, Selina Chen, Dang Nguyen, Ti Lam, Rose Hanna, Lisa Fish, Magdalena Gebala, Alexx J. Smith, Sukh Sekhon, Jennifer Yen, Jeff Gregg, Hani Goodarzi, Helen Li, Fereydoun Hormozdiari, Babak Behsaz, Anna Hartwig, Lee Schwartzberg, Babak Alipanahi. Detection of early stage colorectal cancer using cell-free oncRNA biomarkers and AI [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7133.