Abstract B008: <i>Machine-learning modelling of lung cancer metastasis to the brain using alterations in DNA methylation</i>
Bibliographic record
Abstract
Abstract Background: Brain metastases (BM) are common and arise in 30% of lung adenocarcinoma (LUAD) patients, the most common cancer type to metastasize to the brain. Patients with LUAD that develop BM experience significantly poorer outcomes, with a 10-16 month median overall survival. Unfortunately, current clinical practice for BM prediction is limited and so BM are typically detected after they develop and grow to cause neurological symptoms. Once BM are detected, currently neurosurgical tumor biopsies are performed to enable BM diagnosis via neuropathological evaluation. The objective of this study was to develop DNA methylation-based models that 1) predict which LUAD patients are likely to develop BM and 2) detect BM through a liquid biopsy approach to allow for potential BM prevention, early treatment, and non-invasive diagnosis. Methods: A cohort of 346 LUAD and BM patients was assembled with a combined total of 402 tumor tissue and plasma samples. Machine learning models were built, using discovery datasets (60% and 80%, respectively), with DNA methylation alterations that can stratify the risk of BM development in tissue and can detect BM in plasma. Models were evaluated in independent validation datasets and further validated in additional external data. A predictive nomogram was developed that incorporates the results of the BM prediction model with predictive clinical factors to provide composite patient-specific scores reflecting BM risk. Results: The methylation-based model using LUAD tissue to predict BM was shown to reliably and accurately stratify BM risk in a univariable Cox model using validation set data (hazard ratio [HR]=5.65, 95% confidence interval [CI]: 1.85–17.2, p=0.0023). The utility of this model was independent of the predictive value of clinical factors in a multivariable Cox model using validation set data (HR=8.92, 95% CI: 1.97–40.5, p=0.0046). The BM predictive model had a 5-year area under receiver operating curve (AUROC) of 0.81 which was significantly higher than that of a similarly built model using clinical factors (AUROC=0.65), reflecting its utility over current clinical practice. The predictive nomogram using clinical and methylation-based factors combinatorially had a 5-year BM prediction accuracy of 0.82 and a greater HR in a univariable Cox model (HR=17.2, 95% CI: 4.13–71.3, p<0.0001) in validation set data, demonstrating that it is an optimized patient-specific prediction tool. The methylation-based model for detection of BM in plasma showed accurate classification of BM from gliomas and lymphomas (AUROC=0.80), as typical clinical differential diagnoses, in validation set data. The models were validated further in additional external data. Conclusions: DNA methylation-based modeling of brain metastasis can accurately predict LUAD patients at risk for BM development and can non-invasively detect BM that develop. Future treatment approaches may tailor initial LUAD treatment and ongoing cancer surveillance to a patient’s BM risk, allowing for the potential to prevent and treat BM early. Citation Format: Jeffrey A Zuccato, Yasin Mamatjan, Farshad Nassiri, Andrew Ajisebutu, Jeffrey Liu, Ammara Muazzam, Olivia Singh, Wen Zhang, Mathew Voisin, Suganth Suppiah, Olli Saarela, Ming Tsao, Thomas Kislinger, Kenneth Aldape, Michael Moran, Vikas Patil, Gelareh Zadeh. Machine-learning modelling of lung cancer metastasis to the brain using alterations in DNA methylation [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: DNA Methylation, Clonal Hematopoiesis, and Cancer; 2025 Feb 1-4; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2025;85(3 Suppl):Abstract nr B008.
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 | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
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".