Bias invariant RNA-seq metadata annotation

Abstract

Recent technological advances have resulted in an unprecedented increase in publicly available biomedical data, yet the reuse of the data is often precluded by experimental bias and a lack of annotation depth and consistency. Here we investigate RNA-seq metadata prediction based on gene expression values. We present a deep-learning based domain adaptation algorithm for the automatic annotation of RNA-seq metadata. We show how our algorithm outperforms existing approaches as well as traditional deep learning methods for the prediction of tissue, sample source, and patient sex information across several large data repositories. By using a model architecture similar to siamese networks the algorithm is able to learn biases from datasets with few samples. Our domain adaptation approach achieves metadata annotation accuracies up to 12.3% better than a previously published method. Lastly, we provide a list of more than 10,000 novel tissue and sex label annotations for 8,495 unique SRA samples.

Bias-invariant RNA-sequencing metadata annotation

“…A list of all SRA samples for which the MetaSRA labels and the predicted labels mismatched is available in the Supplementary Material [ 30 ].…”

“…We provide this information such that users can make their own decision on probability cut-offs applied to each class. We provide the full list of all classified SRA samples as well as the probability output of the classifier in the Supplementary Material [ 30 ].…”

“…Previously unlabeled SRA data were automatically annotated with the appropriate metadata. A list of all new metadata can be downloaded with the Supplementary Material [ 30 ].…”

“…For sex classification, genes with 0.4 < g < 0.7 were used ( d = 190). Sample source classification included genes with 0.3 < g < 0.8 (d = 8,679) ( Supplementary Table S2 , list of input features in Supplementary Material [ 30 ]).…”

“…For each of the 2 selected categories we sorted all available studies by number of samples, placed the first third of studies into the training (studies = 420, n = 12,725), the second third into the test (studies = 422, n = 3,144), and the last third into the SRA validation set (studies = 418, n = 1,124) ( Supplementary Tables S2 and S3 ). A list of the sample IDs and corresponding labels is available in the Supplementary Material [ 30 ].…”