Deep learning can predict multi-omic biomarkers from routine pathology images: A systematic large-scale study

Abstract

We assessed the pan-cancer predictability of multi-omic biomarkers from haematoxylin and eosin (H&E)-stained whole slide image (WSI) using deep learning and standard evaluation measures throughout a systematic study. A total of 13,443 deep learning (DL) models predicting 4,481 multi-omic biomarkers across 32 cancer types were trained and validated. The investigated biomarkers included genetic mutations, transcriptomic (mRNA) and proteomic under- and over-expression status, metabolomic pathways, established markers relevant for prognosis, including gene expression signatures, molecular subtypes, clinical outcomes and response to treatment. Overall, we established the general feasibility of predicting multi-omic markers across solid cancer types, where 50% of the models could predict biomarkers with the area under the curve (AUC) of more than 0.633 (with 25% of the models having AUC larger than 0.711). Aggregating across the omic types, our deep learning models achieved the following performance: mean AUC of 0.634 ±0.117 in predicting driver SNV mutations; 0.637 ±0.108 for over-/under-expression of transcriptomic genes; 0.666 ±0.108 for over-/under-expression of proteomes; 0.564 ±0.081 for metabolomic pathways; 0.653 ±0.097 for gene signatures and molecular subtypes; 0.742 ±0.120 for standard of care biomarkers; and 0.671 ±0.120 for clinical outcomes and treatment responses. The biomarkers were shown to be detectable from routine histology images across all investigated cancer types, with aggregate mean AUC exceeding 0.62 in almost all cancers. In addition, we observed that predictability is reproducible within-marker and less dependent on sample size and positivity ratio, indicating a degree of true predictability inherent to the biomarker itself.

Self-supervised attention-based deep learning for pan-cancer mutation prediction from histopathology

“…Initial studies demonstrated this predictability in lung cancer 3 , breast cancer 4 , and colorectal cancer 5 . Subsequently, several "pan-cancer" studies showed that DL-based prediction of biomarkers is feasible across the whole spectrum of human cancer [6][7][8][9][10] . However, these studies were overwhelmingly performed in a single large cohort without externally validating the results on a large scale.…”

Machine learning enabled prediction of digital biomarkers from whole slide histopathology images

“…A broad comparison to the results of Arslan et al is challenging due to the very limited overlap in biomarkers between their work and ours. However, for CDK4, which is the only shared RNA biomarker, they report at AUROC of around 0.72 (averaged across 3 cancer types) [16], whereas we obtain an AUROC of 0.84 pan-cancer; of 0.75 in a stratified analysis, averaged across all cancer types; and of 0.77 when filtering to 4 relevant cancer types (specifically, breast, colorectal, lung, and pancreatic).…”

“…A broad comparison to the results of Arslan et al is challenging due to the very limited overlap in biomarkers between their work and ours. However, for CDK4, which is the only shared RNA biomarker, they report at AUROC of around 0.72 (averaged across 3 cancer types) [16],…”

Latent transcriptional programs reveal histology-encoded tumor features spanning tissue origins

“…Recent studies have shown that machine learning methods can use tumor histology to predict key disease characteristics, including grade [36][37][38] , immune infiltration [39][40][41] , HPV status 42 , mutation status 19,[43][44][45] , expression of individual genes 20,46,47 , and established multi-omic features 41 , but have not attempted to predict de novo learned, complex transcriptional features, such as our CLFs. To assess whether and how each CLF was histologically encoded, we trained a deep learning network to predict binarized CLF status (high or low) from tumor histology images (Fig.…”