Back

A deep learning approach for improved detection of homologous recombination deficiency from shallow genomic profiles

Andre, G.; Coletta, T.; Pozzorini, C.; Marques, A. C.; Bieler, J.; Kempfer, R.; Chong, C.; Saitta, A.; Smith, E.; Macheret, M.; Janiszewski, A.; Bonilla, X.; Bonet, J.; Santos-Silva, H.; Postl, M.; Wozelka-Oltjan, L.; Arrigo, N.; Willig, A.; Grimm, C.; Mullauer, L.; Xu, Z.

2022-07-07 cancer biology
10.1101/2022.07.06.498851 bioRxiv
Show abstract

Homologous Recombination Deficiency (HRD) is a predictive biomarker of poly-ADP ribose polymerase 1 inhibitors (PARPi) response. Most HRD detection methods are based on genome wide enumeration of scarring events and require deep genome sequence profiles (> 30x). The cost and workflow-specific biases introduced by these genome profiling methods currently limits clinical adoption of HRD testing. We introduce the Genomic Integrity Index (GII), a Convolutional Neuronal Network, that leverages features from low pass (1x) Whole Genome Sequencing data to distinguish HRD positive and negative samples. In a cohort of 230 ovarian and breast cancer, we found GII supports accurate stratification of samples yielding results that are highly concordant with state-of-the-art HRD detection methods (0.865<AUC<0.996) which require 50x deeper coverage. We conclude that the deep learning framework supporting GII allows accurate detection of HRD from shallow genome profiles, reducing biases and data generation costs making it uniquely suited for clinical applications.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.