Oral Sessions (Day1, Day2, Day3)
Poster Presentations (Day1, Day2, Day3)
Plenary Lectures
- Day 2, June 23（Thu.） 16:50-17:35 Room A (Main Hall)
-
2A-PL-1650 PDF
Ambient Mass Spectrometry and Machine Learning in Disease Diagnosis
Mass spectrometry (MS) provides a wealth of chemical information. In our laboratory we combined MS-based metabolite and lipid profiling with machine learning to different clinical applications. First we use paper spray ionization (PSI) coupled with chip-based field asymmetry ion mobility spectrometry (FAIMS) to rapidly obtained the molecular features of breast core-needle biopsy. The predictive LASSO model gave a total accuracy close to 90% by training with about a hundred of patient breast samples. Later we utilized such strategy for real-time diagnosis of breast core-needle biopsy obtained from NTU Cancer Center. The tissue can be classified as benign or tumor within 10 minutes upon surgery. Secondly we utilized multimodal desorption electrospray ionization (DESI) mass spectrometry imaging (MSI)-microscopy image fusion to boost the spatial resolution. By such predictive molecular imaging we were able to find dozens of more potential lipid biomarkers that can be used to determine the tumor margins. We further utilized machine learning algorithms to explore if the serum metabolic features from the hereditary retinal dystrophies patients can be used to provide diagnostic insight into the early screening. Lastly, we developed a molecular diagnostic veterinary platform using mobile mini-MS, in which the metabolite profiles of chicken stools were utilized to predict if the poultry were infected with parasites.