Abstract 21P
Background
The use of neoadjuvant therapy (NAT) for operable breast cancer (BC) has progressively increased over time and it is today recommended by major guidelines. The achievement of a pathological complete response (pCR) is associated with an improved outcome. As a consequence new strategies are required to early identify patients who will not respond. In this persepctive, metabolomics may represent an innovative technology to identify host related factors correlated with outcome. In this research we evaluate the use metabolomics analysis coupled to artificial intelligence to predict treatment outcome for BC patients undergoing NAT.
Methods
Untargeted metabolomics analysis was performed on serum samples from 66 operable BC patients treated with NAT. Small molecules were extracted from serum, derivatized and then analyzed using bi-dimensional gas chromatography/mass spectrometer (GCxGC-MS). The metabolomics profiling was then evaluated according to response to therapy (pCR versus residual disease). A machine learning approach was implemented with Boruta features selection algorithm combined with genetical algorithm as classifier, on a training set including 28 plasma samples and externally validated on the remaing 12 samples.
Results
Among the 66 enrolled patients, 27 (41%) were HER2 +, 23 (35%) and 16 (24%) were luminal B. Overall, 52 patients have received surgery so far. pCR was achieved in 29 patients (56%) and residual disease in 23 (44%). A total of 670 small molecules were quantified by untargeted metabolomics analysis; 77 of these resulted differentially expressed (p <0.05 and fold change > 1.3) between patients achieving a pCR or residual disease. A prediction model, combining metabolomic signatures and machine learning, was implemented on 40 metabolomic profiles. With this approach we were able to correctly identify the type of response with an accuracy of 98%.
Conclusions
By using this omic approach, we were able to identify a metabolomic signature correlated with the type of response to NAT. Among differentially expressed molecules, we identified several fatty acids, amino acids and small molecules that could be targeted by selected dietary supplements. Updated analysis of the different biological BC subtypes will be presented.
Legal entity responsible for the study
The authors.
Funding
AIRC.
Disclosure
All authors have declared no conflicts of interest.