Abstract B066: An AI approach to unraveling treatment response in pancreatic cancer: Insights from the COMPASS trial leveraging large language models (LLMs)
Bibliographic record
Abstract
Abstract Pancreatic cancer, often termed the “silent killer” due to its vague symptoms and late diagnosis, continues to challenge oncologists with its significant treatment challenges and poor prognosis. Standard-of-care regimens, such as FOLFIRINOX (FFX) and Gemcitabine + nab-paclitaxel (GnP), show varied efficacy among patients. Understanding the factors that contribute to these varied responses is crucial for advancing personalized treatment strategies. This study uses NetraAI, a novel machine learning (ML) approach to analyze whole genome sequencing (WGS) and RNA sequencing (RNASeq) data from pancreatic ductal adenocarcinoma (PDAC) subjects (n=208) prior to FFX or GnP first-line combination therapy from the COMPASS trial, aiming to identify key variables that distinguish between responses to these two therapeutic regimens. NetraAI identifies explainable subpopulations within clinical trial data defined by a set of driving variables (i.e., hypotheses) related to drug response. This approach focuses on explaining only subjects that collectively represent a real effect and does not forcibly explain everyone, to avoid overfitting. Our analysis identified several key factors significantly correlating with treatment response. Downregulated LRRC8E and upregulated PTPRH showed favourable response to GnP, while upregulated UTG1A13p was associated with FFX response. Across all patients (stable disease, partial response, complete response) CLEC19A and LRRC29 expression were associated with better response to both FFX and GnP, suggesting that they may play a role in the chemotherapy response mechanisms. The NetraGPT module, leverages large language models (LLMs) and NetraAI insights to further query trial results. Querying NetraGPT, we identified a link between LRRC29 and HOOK1, associated with SHP, which has been linked to GnP response in non-small cell lung cancer. These findings suggest that chemotherapy response may be cancer-agnostic, depending on the chemotherapeutic agent, with LRRC29 expression discriminating selecting GnP versus FFX treatment. Our findings highlight the potential of AI in elucidating complex relationships within clinical trial data, offering valuable insights into the determinants of treatment response in pancreatic cancer to achieve more personalized cancer care. The identified biomarkers provide a promising foundation for developing personalized treatment strategies, enhancing the precision of therapeutic decisions. Integrating explainable ML methods for small and medium sized clinical data with LLMs provides an exciting step forward. Retrieval Augmentation allows researchers to create LLMs that are designed to answer questions about heterogeneous patient populations through the vantage of the various insights that have been discovered via specialized ML methods such as NetraAI and its NetraGPT module. We will provide a demo of how an interactive LLM was used to derive new insights into PDAC that may pave the way to precision treatment and explain how researchers can begin to create their own versions of these tools. Citation Format: Joseph Geraci, Bessi Qorri, Mike Tsay, Christian Cumbaa, Paul Leonchyk, Larry Alphs, Luca Pani. An AI approach to unraveling treatment response in pancreatic cancer: Insights from the COMPASS trial leveraging large language models (LLMs) [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Advances in Pancreatic Cancer Research; 2024 Sep 15-18; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2024;84(17 Suppl_2):Abstract nr B066.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.010 | 0.012 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".