Abstract 147P
Background
Non-small cell lung cancer (NSCLC) is one of the most prevalent cancers worldwide. Centre Léon Bérard, Lyon, France has developed a 27-gene expression-based HOT score that predicts outcome in patients with advanced stage treated with immunotherapy. Yet, this signature could be complemented with additional models involving a smaller set of genes. In this study, we used KEM® (Knowledge Extraction and Management) explainable Artificial Intelligence platform, a tool that systematically extracts relations within the variables of a database, to refine the HOT score, shortening the signature and improving performances.
Methods
Starting from GEO warehouse (GSE161537), 2,568 variables were aggregated into a consolidated database with 82 NSCLC patients. Analysis focused on the associations between combinations of genes, previous treatment lines and overall survival (OS): KEM® extracted 194,349 relations that were filtered on the number of involved genes, statistical significance and specificity.
Results
We identified 5 genes, DKC1, HPGD, MLPH, ABCC4 and MVP, combined into two 4-gene signatures with similar performances that predicted survival with a significant interaction between previous treatment lines and gene expression: patients who carry these signatures and that had undergone at least two treatment lines before immunotherapy showed an improved OS in both models (hazard ratio = 0.45 and 0.36). Compared to the HOT score, the number of genes retained was reduced from 27 to 4, and performances improved: balanced accuracy increases from 0.62 (HOT score) to 0.78.
Conclusions
Our analysis enabled the preliminary identification of signatures that complement the HOT score: the impact of previous lines of therapy was included into the model, the number of markers was reduced and performances increased. These findings are currently being confirmed using Cancer Research Institute iAtlas, a database that contains gene expression for over 1,100 cancer patients across five different tissue types. Specific adjustments may be required to account for differences in gene expression measurements between the training and testing tests.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
Has not received any funding.
Disclosure
All authors have declared no conflicts of interest.
Resources from the same session
166P - Metabolomic prediction of breast cancer treatment toxicities
Presenter: Max Piffoux
Session: Poster session 01
167P - A tumor immune microenvironment-based model for prediction of everolimus efficacy in premenopausal women with hormone receptor-positive/human epidermal growth factor receptor 2-negative advanced breast cancer: Preliminary results from MIRACLE trial
Presenter: Tan Yujing
Session: Poster session 01
168P - HUWE1 inhibition has tumor suppressive effect in triple-negative breast cancer cell lines by modulating glycolytic and immune modulatory markers
Presenter: SHRUTI KAHOL
Session: Poster session 01
169P - Integration of metabolomics and transcriptomics to reveal potential biomarkers associated with treatment response of neoadjuvant therapy in HER2+ breast cancer
Presenter: Ningning Zhang
Session: Poster session 01
170P - Clinical significance and functional role of GPR56 (ADGRG1) in breast cancer
Presenter: Haizhu Chen
Session: Poster session 01
172P - T cell-derived circulating DNA and tumour inflammatory microenvironment in EGFR-mutant advanced non-small cell lung cancer: Correlation with the outcome of EGFR TKI treatment
Presenter: Nicha Zungsontiporn
Session: Poster session 01
173P - Expression of programmed death-ligand 1 and EGFR on circulating tumour cells in advanced lung cancer patients
Presenter: Jayant Khandare
Session: Poster session 01
174P - Frequency and prognostic value of circulating tumor cells in cancer of unknown primary
Presenter: Maria Pouyiourou
Session: Poster session 01
175P - Radiomic biomarker of vessel tortuosity for monitoring treatment change: Preliminary findings in prospective evaluation of ECOG-ACRIN EA5163
Presenter: Pushkar Mutha
Session: Poster session 01