Journal: Molecular Oncology
Article Title: A large‐scale retrospective study in metastatic breast cancer patients using circulating tumour DNA and machine learning to predict treatment outcome and progression‐free survival
doi: 10.1002/1878-0261.70015
Figure Lengend Snippet: BAY‐ML model. (A) Visual summary of the two‐stage model. In the first stage, the longitudinal set of tumour fraction scores is modelled using the tumour and treatment features using a random effects model. In the second stage, the CT scan is predicted using the tumour fraction trend in the patient from the random effects and the tumour and treatment features. (B) Receiver operator characteristic (ROC) curve of the dynamic predictive model described in the text (ROC = 0.74 for BAY‐ML with ctDNA info, ROC = 0.71 for BAY‐ML without ctDNA info), together with a simple model that only uses the most recent ichorCNA value. Note that the curves have been built using the maximum number of validation samples for each method. In the case of BAY‐ML models, the predictions were obtained with the last CT scan for each patient left out when the model was fit (129 CT scans). In the case of the model with only ichorCNA, 689 CT scans. The square represents our optimal threshold for 66% specificity and 75% sensitivity, using cross‐validation. The circle represents the performance of CA 15‐3 using the recommended threshold on (C) Number of true/false positives and negatives over 100 patients when the simplest threshold model (TM in the legend) and when the longitudinal ctDNA scores are considered or not into the BAY‐ML model (BAY without and BAY with in the legend). (D) Instances where the model correctly predicted progression and instances where it did not, comparing the available information at that moment. Only predictions within 90 days of the CT‐scan and where predictions are possible in both the threshold and BAY‐ML are shown. BAY‐ML, Bayesian machine learning model; CT, computed tomography; ctDNA, circulating tumour DNA; NGTAS, next generation‐targeted amplicon sequencing; TM, thresholding model; CA15‐3, carcinoma antigen 15‐3.
Article Snippet: The Fluidigm Access arrayTM platform (NG‐TAS) [ ] was used to perform a triplicate target sequencing for 377 Amplicons to analyse 20 breast cancer genes.
Techniques: Computed Tomography, Biomarker Discovery, Amplification, Sequencing
fluidigm is a verified supplier 
fluidigm manufactures this product 
