Background: Breast cancer is the most common cancer in women worldwide. Here we aimed to develop an effective non-invasive method to screen for breast cancer and reduce mortality while still being curable. Methods: Here we propose a method that leverages the available data by incorporating information on copy number variations, mutation signature, and fragment size. Our approach adopted principal component analysis and a generalized linear model algorithm to distinguish between breast cancer and normal samples. Results: A total of 100 samples (85 tumor, 15 controls) were used for training, and 44 samples (37 tumor, 7 controls) were used to validate the proposed method based on whether the sample originated from breast cancer. Our model reached an area under the receiver operating characteristic curve reached 1.0 and 0.690 in the training set and in the validation set, respectively. Conclusions: Our method can differentiate between breast cancer patients and controls using non-invasive, cost-effective, low-coverage whole-genome sequencing technology that may provide new ideas for future breast cancer screenings.