Cancer classification based on cancer signals originating from thresholding.

Abstract

Methods and systems for detecting cancer and / or determining the origin of cancerous tissue are disclosed. In some embodiments, a multiclass cancer classifier trained on multiple biological samples containing cfDNA fragments is disclosed. The analysis system derives a feature vector for each sample, and the multiclass classifier predicts the stochastic likelihood for each of multiple cancer signal origin (CSO) classes. In some embodiments, the multiple CSO classes include hematological subtypes, including both hematological malignancies and progenitor states. In one embodiment, non-cancer samples with high predictive scores are excluded from the training sample set. In another embodiment, the analysis system stratifies the samples according to their predictive scores and applies a binary threshold cutoff determined for each stratum.

Images