Deep learning describes a class of machine learning algorithms that are capable of combining raw inputs into layers of intermediate features. These algorithms have recently shown impressive results across a variety of domains. Biology and medicine are data-rich disciplines, but the data are complex and often ill-understood. Hence, deep learning techniques may be particularly well suited to solve problems of these fields. We examine applications of deep learning to a variety of biomedical problems—patient classification, fundamental biological processes and treatment of patients—and discuss whether deep learning will be able to transform these tasks or if the biomedical sphere poses unique challenges. Following from an extensive literature review, we find that deep learning has yet to revolutionize biomedicine or definitively resolve any of the most pressing challenges in the field, but promising advances have been made on the prior state of the art. Even though improvements over previous baselines have been modest in general, the recent progress indicates that deep learning methods will provide valuable means for speeding up or aiding human investigation. Though progress has been made linking a specific neural network's prediction to input features, understanding how users should interpret these models to make testable hypotheses about the system under study remains an open challenge. Furthermore, the limited amount of labelled data for training presents problems in some domains, as do legal and privacy constraints on work with sensitive health records. Nonetheless, we foresee deep learning enabling changes at both bench and bedside with the potential to transform several areas of biology and medicine.

Inflammation is a beneficial host response to infection but can contribute to inflammatory disease if unregulated. The TH17 lineage of T helper (TH) cells can cause severe human inflammatory diseases. These cells exhibit both instability (they can cease to express their signature cytokine, IL-17A)1 and plasticity (they can start expressing cytokines typical of other lineages)1,2 upon in vitro re-stimulation. However, technical limitations have prevented the transcriptional profiling of pre- and post-conversion TH17 cells ex vivo during immune responses. Thus, it is unknown whether TH17 cell plasticity merely reflects change in expression of a few cytokines, or if TH17 cells physiologically undergo global genetic reprogramming driving their conversion from one T helper cell type to another, a process known as transdifferentiation3,4. Furthermore, although TH17 cell instability/plasticity has been associated with pathogenicity1,2,5, it is unknown whether this could present a therapeutic opportunity, whereby formerly pathogenic TH17 cells could adopt an anti-inflammatory fate. Here we used two new fate-mapping mouse models to track TH17 cells during immune responses to show that CD4+ T cells that formerly expressed IL-17A go on to acquire an anti-inflammatory phenotype. The transdifferentiation of TH17 into regulatory T cells was illustrated by a change in their signature transcriptional profile and the acquisition of potent regulatory capacity. Comparisons of the transcriptional profiles of pre- and postconversion TH17 cells also revealed a role for canonical TGF-β signalling and consequently for the aryl hydrocarbon receptor (AhR) in conversion. Thus, TH17 cells transdifferentiate into regulatory cells, and contribute to the resolution of inflammation. Our data suggest that TH17 cell instability and plasticity is a therapeutic opportunity for inflammatory diseases.

Deep learning, which describes a class of machine learning algorithms, has recently showed impressive results across a variety of domains. Biology and medicine are data rich, but the data are complex and often ill-understood. Problems of this nature may be particularly well-suited to deep learning techniques. We examine applications of deep learning to a variety of biomedical problems—patient classification, fundamental biological processes, and treatment of patients—and discuss whether deep learning will transform these tasks or if the biomedical sphere poses unique challenges. We find that deep learning has yet to revolutionize or definitively resolve any of these problems, but promising advances have been made on the prior state of the art. Even when improvement over a previous baseline has been modest, we have seen signs that deep learning methods may speed or aid human investigation. More work is needed to address concerns related to interpretability and how to best model each problem. Furthermore, the limited amount of labeled data for training presents problems in some domains, as do legal and privacy constraints on work with sensitive health records. Nonetheless, we foresee deep learning powering changes at both bench and bedside with the potential to transform several areas of biology and medicine.

Artificial neural networks (ANN) are computing architectures with many interconnections of simple neural-inspired computing elements, and have been applied to biomedical fields such as imaging analysis and diagnosis. We have developed a new ANN framework called Cox-nnet to predict patient prognosis from high throughput transcriptomics data. In 10 TCGA RNA-Seq data sets, Cox-nnet achieves the same or better predictive accuracy compared to other methods, including Cox-proportional hazards regression (with LASSO, ridge, and mimimax concave penalty), Random Forests Survival and CoxBoost. Cox-nnet also reveals richer biological information, at both the pathway and gene levels. The outputs from the hidden layer node provide an alternative approach for survival-sensitive dimension reduction. In summary, we have developed a new method for accurate and efficient prognosis prediction on high throughput data, with functional biological insights. The source code is freely available at https://github.com/lanagarmire/cox-nnet.

It is crucial for researchers to optimize RNA-seq experimental designs for differential expression detection. Currently, the field lacks general methods to estimate power and sample size for RNA-Seq in complex experimental designs, under the assumption of the negative binomial distribution. We simulate RNA-Seq count data based on parameters estimated from six widely different public data sets (including cell line comparison, tissue comparison, and cancer data sets) and calculate the statistical power in paired and unpaired sample experiments. We comprehensively compare five differential expression analysis packages (DESeq, edgeR, DESeq2, sSeq, and EBSeq) and evaluate their performance by power, receiver operator characteristic (ROC) curves, and other metrics including areas under the curve (AUC), Matthews correlation coefficient (MCC), and F-measures. DESeq2 and edgeR tend to give the best performance in general. Increasing sample size or sequencing depth increases power; however, increasing sample size is more potent than sequencing depth to increase power, especially when the sequencing depth reaches 20 million reads. Long intergenic noncoding RNAs (lincRNA) yields lower power relative to the protein coding mRNAs, given their lower expression level in the same RNA-Seq experiment. On the other hand, paired-sample RNA-Seq significantly enhances the statistical power, confirming the importance of considering the multifactor experimental design. Finally, a local optimal power is achievable for a given budget constraint, and the dominant contributing factor is sample size rather than the sequencing depth. In conclusion, we provide a power analysis tool (http://www2.hawaii.edu/~lgarmire/ RNASeqPowerCalculator.htm) that captures the dispersion in the data and can serve as a practical reference under the budget constraint of RNA-Seq experiments.

Breast cancer is the most common malignancy in women worldwide. With the increasing awareness of heterogeneity in breast cancers, better prediction of breast cancer prognosis is much needed for more personalized treatment and disease management. Towards this goal, we have developed a novel computational model for breast cancer prognosis by combining the Pathway Deregulation Score (PDS) based pathifier algorithm, Cox regression and L1-LASSO penalization method. We trained the model on a set of 236 patients with gene expression data and clinical information, and validated the performance on three diversified testing data sets of 606 patients. To evaluate the performance of the model, we conducted survival analysis of the dichotomized groups, and compared the areas under the curve based on the binary classification. The resulting prognosis genomic model is composed of fifteen pathways (e.g. P53 pathway) that had previously reported cancer relevance, and it successfully differentiated relapse in the training set (log rank p-value = 6.25e-12) and three testing data sets (log rank p-value<0.0005). Moreover, the pathway-based genomic models consistently performed better than gene-based models on all four data sets. We also find strong evidence that combining genomic information with clinical information improved the p-values of prognosis prediction by at least three orders of magnitude in comparison to using either genomic or clinical information alone. In summary, we propose a novel prognosis model that harnesses the pathway-based dysregulation as well as valuable clinical information. The selected pathways in our prognosis model are promising targets for therapeutic intervention.

Clinical challenges exist in reducing prostate cancer (PCa) disparities. The RNA splicing landscape of PCa across racial populations has not been fully explored as a potential molecular mechanism contributing to race-related tumour aggressiveness. Here, we identify novel genome-wide, race-specific RNA splicing events as critical drivers of PCa aggressiveness and therapeutic resistance in African American (AA) men. AA-enriched splice variants of PIK3CD, FGFR3, TSC2 and RASGRP2 contribute to greater oncogenic potential compared with corresponding European American (EA)-expressing variants. Ectopic overexpression of the newly cloned AA-enriched variant, PIK3CD-S, in EA PCa cell lines enhances AKT/mTOR signalling and increases proliferative and invasive capacity in vitro and confers resistance to selective PI3Kδ inhibitor, CAL-101 (idelalisib), in mouse xenograft models. High PIK3CD-S expression in PCa specimens associates with poor survival. These results highlight the potential of RNA splice variants to serve as novel biomarkers and molecular targets for developmental therapeutics in aggressive PCa.

The emerging single-cell RNA-Seq (scRNA-Seq) technology holds the promise to revolutionize our understanding of diseases and associated biological processes at an unprecedented resolution. It opens the door to reveal intercellular heterogeneity and has been employed to a variety of applications, ranging from characterizing cancer cells subpopulations to elucidating tumor resistance mechanisms. Parallel to improving experimental protocols to deal with technological issues, deriving new analytical methods to interpret the complexity in scRNA-Seq data is just as challenging. Here, we review current state-of-the-art bioinformatics tools and methods for scRNA-Seq analysis, as well as addressing some critical analytical challenges that the field faces.