Abstract 1209P
Background
Poly (ADP-ribose) polymerase inhibitors (PARPi) are revolutionary drugs, effective in treating ovarian, breast, pancreatic and prostate cancers. Multiple randomised clinical trials demonstrated efficacy of different drugs from PARPi family, such as Olaparib, Niraparib, Rucaparib and Talazoparib. Despite high efficiency, only a subset of patients truly benefit from the therapy. Some of them carry BRCA1/2 genes mutations, an excellent indicator of Homologous Recombination Deficiency (HRD), the best understood mechanism leading to PARPi sensitivity. However, clinical trials have demonstrated that patients who don’t carry BRCA mutations can also benefit from PARPi. For example, the PRIMA trial showed, that assessing HRD status can channel patients with no BRCA1/2 mutations into successful PARPi treatment. The PRIMA trial also showed that significant proportion of patients without BRCA1/2 mutation or HRD also respond to PARPi. This indicates that current diagnostics for PARPi is not efficient for patient selection.
Methods
By using Artificial Intelligence (AI) and Machine Learning (ML) algorithms, we developed a tool allowing for the precise selection of patients benefiting from PARPi. AI and ML were trained on whole-genome sequencing data from The International Cancer Genome Consortium. Patients were retrospectively and virtually stratified into responders and no-responders. Kaplan–Meier estimate was used to calculate Median Overall Survival (MOS).
Results
The algorithm (_ALICE - ArtificiaL Intelligence ClassifiEr) showed superior performance in patient stratification when compared with HRD assessment method (MyChoiceÒ). By performing retrospective analysis of 102 high-grade serous ovarian carcinomas, we observed that _Alice outperformed the best currently available diagnostics by increasing MOS by 18 months (7 vs 25 months). This substantial improvement is a result of analysing genomic features beyond BRCA status and HRD.
Conclusions
In conclusion, _ALICE has potential to become a leading diagnostic companion tool for PARPi treatment and currently is being optimised for prospective observational clinical trial.
Clinical trial identification
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
MNM Diagnostics.
Disclosure
P. Zawadzki: Shareholder/Stockholder/Stock options, I am a CEO of University start-up involved in presented work: MNM Diagnostics. A. Piotrowska: Research grant/Funding (institution): MNM Diagnostics. M. Wojtaszewska: Full/Part-time employment: MNM Diagnostics. P. Sztromwasser: Full/Part-time employment: MNM Diagnostics.