DeepCNA: an explainable deep learning method for cancer diagnosis and cancer-specific patterns of copy number aberrations

Data associated with the manuscript: https://www.biorxiv.org/content/10.1101/2024.03.08.584160v1

A deep learning approach classified samples according to cancer primary site, based only on binned copy number aberations. Guided integrated gradients was then used to calculate an attribute score for each bin, representing how important that genomic location was in the cancer classification.

Output of the analysis:

• Binned copy numbers for 6707 samples from the Genomics England Limited data set (channel 0 = total copy number, channel 1 = minor allele copy number)
• Attribute scores for how important each genomic bin was for the cancer classification (channel 0 and channel 1)
• Metadata linking rows in the binned data above to sample information

Input data for the analysis

• DATA_CN.tgz contains each sample's copy number profile as an .npy file
• DATA.tgz contains each sample's normalised copy number profile as an .npy file. This is the input to the neural network training in the DeepCNA repository