Pancreatic ductal adenocarcinoma (PDAC) suffers from a lack of an effective diagnostic method, which hampers improvement in patient survival. Carbohydrate antigen 19-9 (CA19–9) is the only FDA-approved blood biomarker for PDAC, yet its clinical utility is limited due to suboptimal performance. Liquid chromatography-mass spectrometry (LC-MS) has emerged as a burgeoning technology in clinical proteomics for the discovery, verification, and validation of novel biomarkers. A plethora of protein biomarker candidates for PDAC have been identified using LC-MS, yet few has successfully transitioned into clinical practice. This translational standstill is owed partly to insufficient considerations of practical needs and perspectives of clinical implementation during biomarker development pipelines, such as demonstrating the analytical robustness of proposed biomarkers which is critical for transitioning from research-grade to clinical-grade assays. Moreover, the throughput and cost-effectiveness of proposed assays ought to be considered concomitantly from the early phases of the biomarker pipelines for enhancing widespread adoption in clinical settings. Here, we developed a fit-for-purpose multi-marker panel for PDAC diagnosis by consolidating analytically robust biomarkers as well as employing a relatively simple LC-MS protocol. In the discovery phase, we comprehensively surveyed putative PDAC biomarkers from both in-house data and prior studies. In the verification phase, we developed a multiple-reaction monitoring (MRM)-MS-based proteomic assay using surrogate peptides that passed stringent analytical validation tests. We adopted a high-throughput protocol including a short gradient (<10 min) and simple sample preparation (no depletion or enrichment steps). Additionally, we developed our assay using serum samples, which are usually the preferred biospecimen in clinical settings. We developed predictive models based on our final panel of 12 protein biomarkers combined with CA19-9, which showed improved diagnostic performance compared to using CA19-9 alone in discriminating PDAC from non-PDAC controls including healthy individuals and patients with benign pancreatic diseases. A large-scale clinical validation is underway to demonstrate the clinical validity of our novel panel.
Background/Objectives: A newly developed nine-protein serum signature has been utilized to enhance the accuracy of an existing three-protein signature used as a blood-based diagnostic tool. This study used the new nine-protein serum signature to evaluate the clinical sensitivity and specificity of a medical device designed to test the clinical performance of an artificial intelligence algorithm. Methods: A blood-based test using multiple reaction monitoring via mass spectrometry was performed to quantify nine proteins (APOC1, CHL1, FN1, VWF, PPBP, CLU, PRDX6, PRG4, and MMP9) in serum samples from 243 healthy controls and 222 patients with breast cancer. Results: Based on cutoff values determined by an artificial intelligence-based deep learning model, the sensitivity and specificity of the nine-protein signature in diagnosing breast cancer among all participants was 83.3% and 88.1%, respectively, whereas those of the three-protein signature were 71.6% and 85.3%, respectively. The assay yielded a positive predictive value of 86.5% for breast cancer and 13.6% for healthy controls, with corresponding negative predictive values of 14.7% and 85.3%, respectively. The accuracies of nine- and three-protein signatures were 85.8% (area under the receiver operating characteristic curve: 0.8526) and 77.0%, respectively. Conclusions: The nine-protein signature may help detect breast cancer more accurately and effectively than the three-protein signature.
Peptide fragmentation spectra contain critical information for the identification of peptides by mass spectrometry. In this study, we developed an algorithm that more accurately predicts the high-intensity peaks among the peptide spectra. The training data are composed of 180,833 peptides from the National Institute of Standards and Technology and Proteomics Identification database, which were fragmented by either quadrupole time-of-flight or triple-quadrupole collision-induced dissociation methods. Exploratory analysis of the peptide fragmentation pattern was focused on the highest intensity peaks that showed proline, peptide length, and a sliding window of four amino acid combination that can be exploited as key features. The amino acid sequence of each peptide and each of the key features were allocated to different layers of the model, where recurrent neural network, convolutional neural network, and fully connected neural network were used. The trained model, PrAI-frag, accurately predicts the fragmentation spectra compared to previous machine learning-based prediction algorithms. The model excels at high-intensity peak prediction, which is advantageous to selective/multiple reaction monitoring application. PrAI-frag is provided via a Web server which can be used for peptides of length 6-15.